DEV Community: calliope

What Are Full-Well Capacity, Dynamic Range, and Saturated Pixels in Industrial Cameras?

calliope — Wed, 04 Nov 2026 08:51:02 +0000

Introduction: Why Does Dynamic Range Matter in Industrial Imaging? In machine vision inspection, image quality directly affects the accuracy of defect detection, dimensional measurement, object positioning, and automated decision-making. However, even a high-resolution industrial camera may struggle to capture all relevant details under challenging lighting conditions.

For example, specular reflections on metal surfaces can create extremely bright regions, while scratches and dents may appear in darker areas. If a camera cannot capture this range of intensities, highlights may become saturated and shadow details may be lost in noise.

This is why dynamic range is an important consideration when selecting an industrial camera. Understanding full-well capacity, dynamic range, saturated pixels, and bit depth helps engineers choose suitable cameras, configure exposure settings, and optimize machine vision imaging systems.

What Is Full-Well Capacity? 1.1 Definition of Full-Well Capacity Full-well capacity (FWC) refers to the maximum amount of electrical charge that an individual image sensor pixel can store. It is typically measured in electrons (e⁻).

When light reaches a CMOS image sensor, each pixel converts incoming photons into electrons and accumulates electrical charge. As exposure time or light intensity increases, more electrons are collected.

Once a pixel reaches its charge-storage limit, it can no longer effectively record additional incoming light. This limit is known as its full-well capacity.

1.2 Understanding Full-Well Capacity A pixel can be compared to a bucket:

Incoming photons are like water flowing into the bucket.

Accumulated electrons represent the amount of water inside.

Full-well capacity is the bucket's maximum volume.

Once the bucket is full, additional water cannot be accurately recorded.

A larger full-well capacity generally allows a pixel to accommodate stronger light signals before reaching saturation. This can be beneficial in bright-field inspection, reflective-surface imaging, and scenes with substantial intensity variations.

1.3 Relationship Between Full-Well Capacity and Pixel Size Under otherwise similar design conditions, larger pixels generally provide more charge-storage capacity and may achieve higher full-well capacity.

However, pixel size is not the only determining factor. Pixel architecture, sensor fabrication, operating voltage, and readout circuitry also affect actual full-well capacity.

Larger pixels do not automatically guarantee better overall imaging performance. Although they may improve charge collection, fewer pixels may fit within the same sensor area, potentially reducing spatial resolution.

What Is Dynamic Range in an Industrial Camera? 2.1 Definition of Dynamic Range Industrial camera dynamic range describes the ratio between the strongest usable signal and the weakest detectable signal that a camera can record.

In practical terms, dynamic range indicates how effectively a camera can capture both bright regions and dark details within the same image.

2.2 How Is Dynamic Range Calculated? At the sensor level, dynamic range is commonly related to full-well capacity and read noise:

Linear Dynamic Range:

DR = FWC / Read Noise

Dynamic Range in dB:

DRdB = 20 log10(FWC / Read Noise)

Where:

FWC: Full-well capacity, measured in electrons.

Read Noise: Sensor readout noise, measured in electrons.

DR: Dynamic range, expressed as a ratio or in decibels.

For example, if a sensor has a full-well capacity of 20,000 e⁻ and read noise of 5 e⁻:

DR = 20,000 / 5 = 4,000

DRdB = 20 log10(4,000) ≈ 72.0 dB

This means the sensor theoretically covers a signal range of approximately 4000:1.

Actual camera specifications may use different measurement methods, operating modes, or definitions. When comparing camera models, ensure that their dynamic-range values were measured under comparable conditions.

2.3 Why Is High Dynamic Range Important for Industrial Inspection? High dynamic range imaging is particularly useful in the following applications:

Metal Surface Inspection: Bright reflections and fine scratches may appear simultaneously on metal surfaces. A higher dynamic range can help preserve information in both highlights and shadows.

Electronic Component Inspection: During PCB solder-joint, connector, and precision component inspection, localized reflections and shadows can affect defect detection. An appropriate dynamic range helps preserve useful grayscale information.

Transparent Material Inspection: Glass, films, and transparent plastics may contain transmitted light, reflections, and dark defects in the same scene. Insufficient dynamic range can make some regions difficult to image reliably.

High-Speed Motion Inspection: In high-speed inspection, exposure time is often reduced to freeze motion. Short exposures limit the number of photons collected, making sensor noise, illumination intensity, and dynamic range important factors to consider together.

What Are Saturated Pixels, and Why Do They Cause Detail Loss? 3.1 Definition of Saturated Pixels A pixel becomes saturated when the incoming light signal is too strong and reaches the limit of its charge-storage capacity or the camera's signal-processing chain.

Saturated pixels typically appear as pure white areas, clipped highlights, or bright regions with missing texture details.

3.2 Why Does Saturation Cause Detail Loss? Suppose a pixel has a maximum digital output of 4095. If the incoming signal continues to increase after the pixel or ADC reaches its limit, the output remains near the maximum value.

Actual Signal Intensity Output Grayscale Value 3000 3000 3800 3800 4095 4095 5000 4095 6000 4095 Different signal intensities are then mapped to the same output level. The camera can no longer distinguish subtle differences between those regions, resulting in lost highlight detail.

3.3 Saturation and Blooming In some image sensors, excess charge from saturated pixels may spread into neighboring pixels, creating a phenomenon known as blooming.

Blooming may appear as white halos, spreading bright spots, or localized streaks around saturated regions, potentially affecting defect detection and dimensional measurement.

It is important to distinguish saturation from blooming: not every saturated pixel produces noticeable blooming. The severity depends on the sensor architecture and anti-blooming design.

What Is the Difference Between Full-Well Capacity, Dynamic Range, Bit Depth, and SNR? These four parameters are related, but they describe different aspects of camera performance.

Parameter Meaning Imaging Impact Full-Well Capacity Maximum charge stored by a pixel Determines the capacity for strong signals Dynamic Range Ratio between the strongest usable and weakest detectable signals Determines the ability to capture bright and dark details Bit Depth Number of bits used to digitize the signal Determines the number of available digital levels SNR Ratio of signal to noise Affects image clarity, detail recognition, and measurement stability 4.1 Does Higher Bit Depth Always Mean Higher Dynamic Range? Not necessarily. Bit depth determines the number of digital intensity levels:

N = 2n

For example:

8-bit: 256 grayscale levels.

10-bit: 1,024 grayscale levels.

12-bit: 4,096 grayscale levels.

16-bit: 65,536 grayscale levels.

However, if the sensor has high noise or pixels saturate before reaching the ADC's maximum value, increasing bit depth cannot automatically recover lost optical information.

4.2 Relationship Between SNR and Dynamic Range Signal-to-noise ratio (SNR) describes the strength of a useful signal relative to noise. A higher SNR generally makes image information easier to distinguish from noise.

Dynamic range, on the other hand, describes the overall range of signal intensities that the camera can capture.

Both are important to image quality, but they are not interchangeable. A high-dynamic-range camera does not necessarily provide the highest SNR under every exposure condition, and high SNR alone does not guarantee the ability to capture both extremely bright and dark details.

What Factors Affect Industrial Camera Dynamic Range? 5.1 Image Sensor Design Pixel architecture, full-well capacity, read noise, quantum efficiency, and sensor fabrication all influence dynamic range.

5.2 Exposure Time Longer exposure times generally allow pixels to collect more photoelectrons, but they also increase the risk of saturation in bright regions.

5.3 Gain Settings Gain amplifies the sensor output, making dark signals easier to distinguish. However, it also amplifies noise and may reduce available highlight headroom.

5.4 Lighting and Optical System Lighting intensity, illumination angle, lens aperture, filters, and surface reflectivity all affect the amount of light reaching the sensor.

5.5 Image Processing and Output Modes Gamma, LUT, auto-exposure, auto-gain, and image-enhancement algorithms may alter the displayed image.

For quantitative measurement or dynamic-range evaluation, engineers should prioritize raw image data and verify the camera's output bit depth, gain mode, and image-processing settings.

How Can Exposure, Gain, and Lighting Improve Image Quality? 6.1 Set an Appropriate Exposure Time If the image contains obvious saturated highlights, try reducing exposure time first.

If shadow details are insufficient, exposure time can be increased, while considering motion blur and inspection throughput.

6.2 Control Gain In industrial inspection, it is generally preferable to obtain sufficient signal through lighting and exposure adjustments before increasing gain.

Excessive gain may increase noise, compress useful grayscale separation, and reduce measurement stability.

6.3 Optimize Lighting A well-designed lighting system is often more effective than simply increasing camera settings. Common approaches include:

Diffuse lighting to reduce specular reflections.

Polarizing filters to suppress selected reflections.

Backlighting to emphasize contours and dimensions.

Coaxial illumination to improve imaging uniformity on reflective surfaces.

Low-angle, ring, or bar lighting selected according to defect characteristics.

6.4 Use ROI and Suitable Acquisition Modes For high-speed inspection, a region of interest (ROI) can reduce unnecessary data transmission and improve acquisition efficiency.

However, ROI primarily improves acquisition efficiency and frame rate. It does not directly increase the sensor's full-well capacity or intrinsic dynamic range.

MindVision Industrial Cameras for High-Dynamic-Range Imaging In industrial vision systems, camera selection should consider dynamic range, resolution, frame rate, exposure mode, interface, and actual lighting conditions.

MindVision's MV-GE507GC/M and MV-GE508GC/M series can be used in a range of industrial inspection applications. With 5-megapixel resolution, GigE connectivity, and 12-bit output, they are suitable for applications requiring both detailed imaging and industrial system integration.

7.1 MV-GE507GC/M: High-Speed Motion and Precision Inspection The MV-GE507GC/M is a 5-megapixel GigE industrial camera featuring global shutter technology, making it suitable for moving-object imaging and precision inspection.

Parameter Specification Resolution 2448 × 2048 Pixel Size 3.4 μm Sensor Size 2/3" Maximum Frame Rate 24 fps Output Bit Depth 12-bit Dynamic Range 66.9 dB (manufacturer specification) SNR 39.3 dB (manufacturer specification) Exposure Global shutter Interface GigE Global shutter helps reduce geometric distortion when imaging moving objects. Its 12-bit output provides a relatively broad range of grayscale levels for detailed inspection.

For example, in electronic component positioning, mechanical part measurement, and moving-object inspection, appropriate exposure and lighting settings can help reduce highlight saturation while preserving useful grayscale information.

7.2 MV-GE508GC/M: General-Purpose Industrial Vision Inspection The MV-GE508GC/M series also features 5-megapixel resolution and GigE connectivity, making it suitable for electronics manufacturing, surface inspection, dimensional measurement, and automated vision positioning.

Parameter Specification Resolution 2448 × 2048 Pixel Size 3.45 μm Sensor Size 2/3" Maximum Frame Rate 23.5 fps Output Bit Depth 12-bit Dynamic Range 62 dB (manufacturer specification) SNR 42 dB (manufacturer specification) Exposure Frame exposure Interface GigE In reflective or high-contrast inspection environments, the MV-GE508GC/M can be combined with suitable lenses, lighting, and exposure settings to acquire stable image data.

For applications requiring greater tolerance to intensity variations, testing under actual working distances, illumination levels, and surface conditions is recommended. Dynamic range, SNR, and saturation should be evaluated together to determine the final imaging performance.

Camera Selection Tip:

Dynamic range alone cannot determine whether a camera is suitable for a particular application. For high-dynamic-range imaging, evaluate:

Actual dynamic range and noise performance.

Whether highlights become saturated.

Whether dark defects can be reliably detected.

Whether exposure time meets motion-inspection requirements.

Whether output bit depth meets downstream image-analysis needs.

Frequently Asked Questions Q1: Is a Higher Dynamic Range Always Better?

Not always. A higher dynamic range helps capture a wider range of brightness levels, but actual inspection performance also depends on noise, resolution, lighting, exposure time, and target characteristics. Camera selection should be application-specific.

Q2: Does a 12-Bit Industrial Camera Always Have High Dynamic Range?

Not necessarily. A 12-bit camera can output up to 4,096 digital grayscale levels, but its actual dynamic range is also limited by full-well capacity, read noise, and the signal-processing chain.

Q3: Why Do Pure White Areas Appear in an Image?

Pure white regions may result from pixel saturation or the ADC reaching its maximum output value. Try reducing exposure time or gain, adjusting illumination, or controlling reflections.

Q4: Does Increasing Gain Improve Dynamic Range?

Generally, no. Increasing gain can make dark signals more visible, but it may also amplify noise and reduce highlight headroom. Lighting and exposure should be optimized first.

Q5: How Can I Judge Whether an Industrial Camera Is Suitable for High-Dynamic-Range Inspection?

In addition to checking the dynamic-range specification, evaluate full-well capacity, SNR, bit depth, exposure mode, and actual test images. For highly reflective targets, pay particular attention to the proportion of saturated pixels and the visibility of dark defects.

Q6: Can Software Recover Details Already Saturated in the Camera?

Generally, no. Image enhancement, gamma, or LUT adjustments can change the displayed appearance, but they cannot reliably recover signal information already clipped by the sensor or ADC. Saturation should be prevented during image acquisition.

Conclusion: Choose the Right Industrial Camera for Reliable Imaging Full-well capacity, dynamic range, saturated pixels, bit depth, and SNR are important parameters for understanding industrial camera performance.

Full-well capacity determines how much charge a pixel can store. Dynamic range describes the signal range a camera can capture. Bit depth determines the number of digital grayscale levels, while SNR affects the distinction between useful image information and noise.

In practical machine vision applications, engineers should combine camera performance, exposure settings, gain control, lighting design, and target characteristics to establish a stable and repeatable imaging solution.

MindVision offers a range of industrial cameras designed for different resolutions, interfaces, and imaging requirements. For applications requiring detailed imaging, motion capture, and industrial integration, explore the MV-GE507GC/M and MV-GE508GC/M series and evaluate them under your actual inspection conditions.

Reference: https://www.mindvision.ltd/

The Automotive "Nervous System" Is Being Reinvented A Look at Wiring Harness Evolution and Visual Inspection

calliope — Tue, 27 Oct 2026 07:51:05 +0000

If the engine is the heart of a car, the wiring harness is its nervous system. This metaphor has been used for years, but few truly realize that this "nervous system" is undergoing an unprecedented transformation.

A single wire is becoming completely different. Its materials are changing, its length is changing, and its mission is changing. Today, we start with the industrial evolution of automotive wiring harnesses, then move into harness inspection, visual inspection, and quality control at the "last centimeter" — connectors and terminals.

Automotive Harness · Visual Inspection · Connectors · Terminals · MindVision

A Hidden Billion-Dollar Track Global automotive wiring harness sales were about USD 71 billion in 2025, expected to reach USD 75.1 billion in 2026, and potentially exceed USD 108.9 billion by 2032, with a CAGR of about 6.4%. Under another estimate, the global automotive harness market is about USD 57.9 billion in 2026 and will grow to USD 92.8 billion by 2034. Definitions differ, but the direction is clear: this is a steadily expanding billion-dollar market.

A fascinating feature of the harness industry is that it lacks the glamour of batteries or chips, yet every vehicle cannot do without it. A traditional ICE vehicle carries about RMB 2,500–4,500 of harness value per unit, while a new energy vehicle averages around RMB 5,000 — nearly double. The reason is simple: EVs add high-voltage connection systems among battery, motor, and electronic control, greatly increasing harness complexity and technical barriers.

Globally, Japanese, American, and European Tier 1 giants still dominate. Yazaki, Sumitomo Electric, and Aptiv together hold about 71% of the global market. In China, however, local players are accelerating. Tianhai Electric, for example, reached 8.45% domestic market share in automotive harnesses in 2024, ranking first in China, with connector share also among the top three domestically.

Aluminum Replacing Copper: The Two-Century "Copper Era" Begins to Loosen If any change best represents the "underlying revolution" in the harness industry, it is the shift in materials.

For nearly two centuries, copper has been the absolute protagonist of automotive wiring. The reasons are solid: excellent conductivity, high mechanical strength, and mature processing. But copper has an increasingly fatal weakness — cost.

LME data shows copper prices surged to a record high of EUR 13,160 per tonne in the first half of 2026, and remain around EUR 11,663 per tonne. Aluminum spot prices are only about EUR 2,719 per tonne — copper is more than four times as expensive. This huge gap is forcing automakers to reconsider material choices.

Ferrari has introduced aluminum high-voltage harnesses in the 296 hybrid and is extending them to its first EV, Luce. BMW's latest e-drive platform uses substantial aluminum wiring in both high- and low-voltage systems. Stellantis has also begun replacing copper with aluminum in some models.

The benefits are clear. Lear's measurements show weight dropping from 21.2 g/m to 11.7 g/m — a 45% reduction. Although aluminum has lower conductivity and needs about 30% more cross-section for the same current capacity, it still enables about 30% lightweighting and significantly lower material cost. Vehicle-wide aluminum harnesses can cut overall cost by about 20%.

But aluminum replacement is not simply swapping materials. Aluminum conductivity is only about 60% of copper, mechanical strength about one-third, and it suffers from creep and oxidation. Joining aluminum to copper terminals is a major pain point: thermal expansion mismatch can create voids at welded interfaces, and aluminum oxide films increase resistance. Ultrasonic welding is currently viable, but the process window is narrow, and the industry is still working on it.

In the short term, aluminum harnesses will first expand in high-voltage systems. Total harness weight can drop by at least 20%, and vehicle weight by over 10 kg. Full copper replacement will take time, but the direction is clear.

Architecture Revolution: From 5 Kilometers to 100 Meters If aluminum replacing copper changes the material, the E/E architecture revolution changes the very way harnesses exist.

Under traditional distributed architectures, every function had its own ECU — engine management, braking, windows — leading to harness lengths of over 5 km and thousands of connectors, jokingly called "spaghetti" wiring. Domain-centralized architectures cut this to about 3 km. The latest "central computing + zonal control" architecture is pushing it toward the hundred-meter level.

Tesla is the most representative: Model S about 5 km, Model 3 about 1.5 km, Cybertruck down to 500 m. Leapmotor's "Four-Leaf Clover" architecture compresses the harness to 996 m, reduces ECUs to 22, and triples system response speed. Hefei Yitong Electronics' LNN system goes further, simplifying harnesses by 70% and cutting assembly time by 85%.

The core logic of zonal architecture is "local access" — sensors and actuators connect to nearby zonal controllers by physical location, not by function. Power distribution is also restructured: secondary power distribution moves into zonal controllers, shifting from "harness-driven" to "node-driven."

Meanwhile, 48V systems are further slimming harnesses. Tesla Cybertruck has already adopted 48V. At the same power, current drops to one quarter, and conductor cross-section can shrink by about 75%.

When Harnesses Begin to "Speak": The Rise of High-Speed Data Harnesses Another change brought by intelligence: harnesses must not only transmit power, but also data.

With ADAS, smart cockpits, and connected vehicles, in-vehicle data demand is growing exponentially. Traditional CAN bus runs at hundreds of Kbps to 1 Mbps, while automotive Ethernet supports 100 Mbps to 10 Gbps — tens to hundreds of times faster. 10 Gbps Ethernet has moved from future technology to a smart-driving necessity.

This imposes new requirements: better shielding, lower signal loss, stricter EMC. Traditional harness design is no longer enough. Companies must upgrade materials, twisting processes, and connector design. Backbone connectors between modules now carry both high-voltage power and high-speed data, pushing connectors toward miniaturization and integration.

Smart Manufacturing: One Harness Every 4 Seconds Another easily overlooked change: production methods are being redefined.

At Langfang Leoni Wiring Systems, smart equipment runs at high speed, producing one finished harness every 4 seconds on average. The company has 31 Mercedes-Benz lines and 18 Xiaomi lines, with annual capacity of about 720,000 sets.

Wuhan's new Vitesco plant uses fully automatic insertion equipment, modular assembly lines, and automatic ultrasonic welding, supported by overhead crane systems and AGVs, achieving full-process automation from raw material to finished product. Its self-developed visual inspection precision exceeds millimeter level.

In Xiaogan High-Tech Zone, a complete harness supply chain is forming — from copper rod drawing at Hubei Chuangxin Electric, to cable processing at Xiaogan Foose, to full vehicle harness assemblies. The local system covers "cable materials — precision harness processing — connector matching — full vehicle harness assembly." In 2025, the zone's harness output reached about RMB 8 billion, growing 10%.

Visual Inspection: The "Second Pair of Eyes" for Harness Quality Control Harnesses are becoming thinner, shorter, and more complex, posing unprecedented quality control challenges. Traditional manual visual inspection is increasingly unable to keep up with growing variants and precision requirements.

Visual inspection is the core technological path to meet this challenge. It uses industrial cameras and algorithms to "understand" every detail of a harness, completing judgments in milliseconds that once required long, focused human attention.

6.1 Why Harnesses Need to Be "Seen"
Automotive harness manufacturing involves cutting, stripping, crimping, housing insertion, and assembly. Every step can introduce defects: incomplete crimping, insulation damage, wrong wire sequence or color, reversed connector polarity, cable surface scratches, jacket cracks, etc.

Most defects are visible, but manual inspection reliability is dropping sharply. Long periods of high concentration cause visual fatigue, fluctuating miss rates, and inconsistent results. More importantly, modern lines run faster, and harness variants multiply. With dozens of specifications and color combinations, human memory and visual recognition easily fail on high-speed lines.

At the same time, OEMs are tightening quality tolerance. A poorly connected terminal may only fail after tens of thousands of kilometers, and recall costs far exceed the cost of catching defects on the line. This "quality leverage" is turning visual inspection from optional to mandatory.

6.2 What to Inspect: Full-Dimensional Coverage from Surface to Interior
Automotive harness visual inspection is not a single technology, but a cluster of technologies for different inspection objects.

Terminal crimp inspection is one of the most difficult scenarios. Crimping mechanically presses the terminal and wire to form an electrical connection; quality directly determines signal reliability. Visual inspection must judge multiple metrics in a tiny area: crimp length, terminal end dimensions, crimp width, insulation crimp depth, exposed copper, etc.

A team at Soonchunhyang University proposed a Harris corner detection method, extracting six key measurements to judge crimp quality. It achieved average 97% accuracy across all items, with 99.9% and 99.7% for crimp width and core width. Its value lies in relying only on traditional image processing, without large annotated datasets, making it highly deployable for cost-sensitive small and medium lines.

Insulation jacket crack inspection faces another challenge. The jacket is a cylindrical surface. Under ordinary visible light, specular reflection hides fine cracks, and bending/twisting during assembly makes single-view images insufficient. A 2026 patent proposed a dual-domain method combining polarization reflection suppression and flattened geometric modeling, using multi-channel input and cross-domain attention for stable crack recognition and quantitative measurement under complex lighting. Multi-camera approaches are also used: synchronized multi-view image sequences, CNN and optical flow build a 3D center-axis model to precisely locate defect radial positions.

Wire sequence and color inspection is relatively mature. Harnesses contain multiple colored wires, and wrong sequence is one of the most common assembly errors. Semi-automatic machine vision systems can simultaneously verify wire position, connector polarity, and color sequence, supporting linear and circular configurations. The core is an HSV + RGB joint classifier, achieving high-precision color discrimination under controlled lighting.

6.3 How Deep Learning Changes the Game
If traditional machine vision solves "can see," deep learning solves "can see accurately" and "can understand."

The difficulty is the high diversity of defect forms. "Poor crimping" may appear as too shallow, too deep, skewed, exposed copper, excessive insulation insertion, etc. Each has different visual features. Rule-based methods require engineers to manually design features and thresholds for every defect — huge effort and hard to exhaust.

In recent years, lightweight YOLO-based models have become a research hotspot. TCQI-YOLOv5 is designed for terminal crimp quality inspection, integrating C2f, FasterNet, and efficient multi-scale attention. It achieves 98.3% average precision for hard-to-detect defects such as "insulation crimp too shallow," while meeting real-time industrial speed requirements.

Another study proposed ILN-YOLOv8, adding a P2 detection layer and efficient local attention. It reached 96.2% accuracy on a real crimp connector dataset, with parameters and computation reduced by 66.7% and 34.6%. This lightweight nature is critical for edge deployment — models must run in real time on limited compute, not rely on cloud inference.

For wire recognition, YOLO-Slender solves a more basic problem: how to quickly and accurately identify and segment each wire under changing lighting, stacking, and many wire types. With C3k2-RDR and a lightweight segmentation head, it reduces parameters by 28.53% while maintaining accuracy, suitable for industrial devices with limited compute.

6.4 An Easily Overlooked Advance: Few-Shot Learning
The weakness of deep learning is data. Traditionally, a reliable defect model needs thousands or tens of thousands of annotated images. In harness production, some defects are extremely rare — it is hard to collect enough "bad" samples.

FFAD offers a new path. Using a pre-trained vision-language model (CLIP), fixed-position image capture, and learnable positional prompts, it achieves over 90% detection accuracy with only 16 samples per defect class. This means new models or harness variants can be deployed in one or two days, without weeks of sample collection and training.

The practical significance: harness variants are exploding — one model may have dozens of configurations, and differences across models are even greater. If every change required large sample collection and retraining, visual inspection could never keep up with line changeover. Few-shot learning is opening this bottleneck.

6.5 A Key Understanding: Visual Inspection Is Not Electrical Testing
One boundary must be clarified: visual inspection and electrical testing (continuity, withstand voltage, insulation resistance) are complementary, not substitutes.

Visual inspection finds appearance and assembly defects — wrong sequence, abnormal crimp form, jacket damage, incomplete connector mating. It cannot verify whether the conductor is truly connected to the correct terminal, whether insulation remains intact under voltage stress, or whether contact resistance is within spec.

Conversely, a harness passing all electrical tests may still have visible mechanical or assembly defects that can later become failures.

Therefore, modern harness inspection uses a "visual + electrical" double defense: visual inspection catches appearance and assembly defects upstream, electrical testing verifies functional integrity downstream. On highly automated lines, the two are merging — the same station performs imaging and continuity testing, and integrated data enables cross-validation and fast traceability.

Connectors and Terminals: The "Last Centimeter" of Harness Quality In the harness manufacturing chain, connectors and terminals are the "nerve endings." Whether a harness works reliably often depends not on the cable itself, but on the crimp quality of dozens or hundreds of terminals and connectors, PIN coplanarity, and spring contact height precision.

These parts are often millimeter or sub-millimeter scale, yet carry core signal and current functions. A poorly connected terminal may fail after tens of thousands of kilometers; a PIN bent by a fraction of a millimeter can interrupt an entire connector module.

This is the most difficult and critical "last centimeter" — terminal and connector inspection.

The challenges are multi-dimensional:

Precision vs. speed. Crimp width is often only a few millimeters; PIN pitch can be 0.4 mm or finer; defect features are micron-scale. Meanwhile, high-speed lines require cycle times in seconds or milliseconds — full-dimensional PIN inspection in 0.1 seconds demands extreme resolution and frame rate.

Diverse defect types. Terminal and connector defects are complex: bent, tilted, missing, broken pins, uneven plating, out-of-tolerance dimensions, corrosion, poor connection stability. Each has different visual features; some are invisible under visible light.

Material and geometry complexity. Terminals are metallic and highly reflective; specular highlights hide real defects. Connectors are compact, pins occlude each other, and a single view often cannot cover the whole inspection area.

Inline engineering challenge. No matter how accurate in the lab, inspection must fit line cycle time and space constraints. Systems must work with crimping and housing insertion machines, achieving stable imaging and fast judgment in tight spaces.

A mature connector visual inspection system must cover the full chain from imaging to judgment.

At the inspection item level, key metrics include spring contact height, PIN pitch and coplanarity, PIN length and diameter verification, plating quality, and connection stability. These form the basis of connector reliability.

At the imaging solution level, camera, lens, and lighting must match object size and precision. For PIN coplanarity requiring high-precision Z measurement, 3D or multi-angle vision is necessary. For PIN bending, sufficient resolution is needed to capture tiny angular deviations.

At the algorithm and judgment level, AI-based adaptive threshold algorithms are becoming mainstream. Compared with fixed thresholds, they better handle batch variation and lighting changes, maintaining detection rate while reducing false alarms.

At the system integration level, inspection must seamlessly connect with line control for real-time output and automatic sorting, ensuring defects do not flow downstream.

Imaging Selection: Industrial Cameras Are the Starting Point Harness inspection places demanding requirements on industrial cameras, and different processes differ greatly.

Terminal crimping stations need high resolution and global shutter. Crimp areas are typically 2–10 mm, defect features often sub-millimeter, requiring sufficient ground resolution and global exposure to avoid geometric distortion in motion. MindVision's PX series, with global shutter CMOS and up to 10 MP, is a reasonable choice — 4608×2176 resolution with proper magnification presents crimp details clearly enough for stable algorithm judgment.

Harness appearance stations face another problem. A harness may contain a dozen colored wires that overlap, bend, or are partly covered. High color fidelity and dynamic range are needed to distinguish colors and boundaries under uneven lighting. MindVision GigE cameras offer full mono/color configurations and up to 25 MP, leaving room for optical selection in multi-variant, multi-color inspection.

High-speed line inspection balances speed and precision. When harnesses pass at high cycle rates, area-scan frame rate and exposure become bottlenecks. MindVision's MV-XGLC83DM-T4-90 line-scan camera, with 8K resolution and up to 150 kHz line rate, targets large-format, high-speed continuous inspection — keeping images sharp without smear, a prerequisite for defect interception. The PCIe line-scan MV-PEL83DM pushes max line rate to 200 kHz, further compressing the imaging window.

For cross-section analysis and copper-aluminum interface inspection after aluminum substitution, the SWIR camera MV-GEC500I uses an InGaAs detector covering 0.4–1.7 μm, "seeing through" interface features invisible in visible light. For aluminum oxide films and Cu-Al weld voids, multispectral imaging is becoming a direction worth watching.

Choosing a camera is essentially choosing whether it can run stably. There is a saying on harness lines: a solution that works in the lab may not survive the first month on the line.

Camera selection often focuses on pixels and frame rate, but in deployment, long-term stability is often determined by less glamorous factors — interface friendliness, SDK maturity, and compatibility cost with existing line control.

Harness lines have distributed stations; cameras are often inside mechanical structures with limited wiring space and narrow maintenance windows. MindVision GigE cameras support 100 m transmission and PoE, providing real value: one cable for power and data, reducing cables and maintenance complexity. They are fully Vision-compliant, with SDK support for Windows, Linux, ARM Linux, and MAC OS, requiring no extra adaptation for HALCON, VisionPro, and other mainstream platforms.

These details rarely appear prominently in PPTs, but they determine the distance between "accepted" and "still running after three years."

From "can capture" to "captures correctly," harness visual inspection is undergoing an interesting shift. Early needs were "can capture" — replacing eyes for basic sequence and appearance checks. Today's need is "captures correctly" — continuously delivering trustworthy images at smaller defect scales, with more complex materials and faster cycles.

This shift imposes systemic requirements. It is no longer a camera plus lens, but coordination of lighting, optics, camera performance, and mechanical positioning. In crimp inspection, if lighting cannot suppress metal highlights, even a high-resolution camera captures only overexposed areas. In color recognition, if color fidelity has systematic bias, even a fine classifier may misjudge after changing wire suppliers.

MindVision's portfolio includes an interesting approach: the MV-NEH501C autofocus industrial camera supports 35 cm to infinity autofocus and uses a platform design, allowing sensor modules with different resolutions to be customized. This flexibility matters in harnesses — one station may handle different lengths and poses; fixed-focus solutions require mechanical repositioning during changeover, while autofocus reduces that cost.

QA: More About Automotive Harness Visual Inspection Q1: Can visual inspection fully replace manual inspection? Under clear rules, controlled lighting, and coverable defect types, visual inspection can approach full inspection with far better efficiency and consistency. But it cannot fully replace humans. Complex surfaces, rare defects, and fast multi-variant changeovers still need human review and exception handling. The realistic path is "machine full inspection + human review," not a one-size-fits-all replacement.

Q2: What is the relationship between visual inspection and electrical testing?
Complementary, not substitute. Visual inspection handles appearance and assembly — sequence, color, crimp form, jacket damage, connector mating. Electrical testing handles continuity, withstand voltage, insulation resistance. A harness may look perfect but have an internal broken core, or pass electrically but have abnormal crimp form. Modern lines should combine both for cross-validation.

Q3: How precise can terminal crimp visual inspection be?
Under controlled conditions, precision is considerable. Harris corner detection achieves about 97% average accuracy across six measurements — crimp length, terminal end dimensions, crimp width — with some dimensions above 99%. Improved YOLO models reach 98.3% average precision for hard defects like "insulation crimp too shallow." Actual line precision depends heavily on camera resolution, lighting, reflection control, and mechanical positioning.

Q4: Does deep learning inspection always need many defect samples?
Traditionally yes, but not necessarily now. Few-shot learning using pre-trained vision-language models can exceed 90% accuracy with only about 16 samples per defect class. This matters greatly because harness variants are many, changeovers fast, and some defects rare. It allows new models and configurations to be deployed in one or two days instead of weeks.

Q5: How does visual inspection handle many colors, strong reflections, and stacking?
Usually at three levels. First, lighting and optics — polarization, diffusion, multi-angle fill to suppress highlights. Second, algorithms — HSV + RGB classifiers for color, segmentation for stacking. Third, mechanics — fixtures to present the harness in the same pose. For curved jacket cracks, multi-camera or 3D reconstruction is used.

Q6: Can visual inspection keep up with faster line cycles?
Lightweight models are solving this. ILN-YOLOv8 maintains accuracy while reducing parameters and computation by about 66.7% and 34.6%, enabling real-time edge operation. Lines also use parallel stations, triggered capture, and ROI cropping to reduce per-frame computation. Visual inspection is not "capture and compute forever," but a real-time process embedded in the cycle.

Q7: Can visual inspection find internal broken cores or abnormal crimp resistance?
Not directly. Vision sees external form. Internal broken cores, high contact resistance, and internal insulation defects require continuity, micro-resistance, withstand voltage, or even X-ray/ultrasound. Visual inspection's value is catching appearance and assembly defects upstream, not covering every quality dimension.

Q8: Does a new harness require a new inspection system?
Ideally no — just change the recipe. Camera, lens, lighting, and station stay; switch models, thresholds, and inspection regions. Few-shot learning and pre-trained models are compressing changeover from weeks to days or less. If every harness change required new equipment, visual inspection could never scale in this fast-paced industry.

Q9: What metrics matter when selecting visual inspection equipment?
Don't just look at pixels. Look at frame rate vs. cycle, lighting stability, lens depth of field, algorithm platform openness, PLC/MES integration, false positive/negative rates, changeover convenience, and supplier field service. For crimp quality, check whether it closes the loop with cross-section analyzers, pull testers, and crimp force monitors — not an isolated vision device.

Q10: Will aluminum replacing copper bring new inspection challenges?
Yes. Aluminum oxide films, reflection characteristics, and Cu-Al interface visual features differ from pure copper. Aluminum is softer, and crimp deformation differs. Existing models and lighting cannot simply be copied; samples must be recollected and optics/thresholds adjusted. Aluminum substitution propagates into inspection.

Q11: How to calculate ROI for visual inspection?
ROI comes from three areas: reducing manual inspection and recheck costs, lowering recall/claim risk from escapes, and improving first-pass yield and cycle time. Harnesses are safety-related; a poor terminal may fail after tens of thousands of kilometers, and recall cost far exceeds line interception cost. For high-speed lines, visual inspection is not a cost item but a quality lever.

Q12: Where is visual inspection heading next?
From "inspection" to "control." Today it mostly finds defects; tomorrow inspection data will feed back to crimpers, cutters, and strippers for adaptive process adjustment. Edge AI, 3D vision, multispectral imaging, few-shot learning, and digital twins will further converge. When vision can warn of crimp parameter drift before defects appear, harness quality competition enters the next stage.

Final Thoughts
The automotive harness industry has been the "silent majority" for decades — not sexy, not eye-catching, but indispensable infrastructure for every vehicle. Today it stands at the intersection of multiple transformations: materials from copper to aluminum, architecture from distributed to centralized, function from power to power plus data, and production from labor-intensive to smart manufacturing.

Visual inspection is becoming indispensable quality infrastructure in this transformation. Its core value is not simply "replacing human eyes with machines," but moving quality control from post-hoc sampling to real-time full inspection, and from experience-based judgment to data-driven decisions.

When a line deploys dozens of visual inspection nodes, each generating multiple images and measurements per second, that data forms a real-time portrait of crimping process state. When a crimp parameter begins to drift — perhaps due to a slight tolerance change in a terminal batch — the vision system can warn before defects occur. This is the deepest value of visual inspection: not just a quality gate, but an entry point for process digitalization.

In the next decade of the harness industry, whoever moves visual inspection from "inspection" to "control" will hold the core advantage of quality competition. And the starting point is letting the camera capture a sufficiently informative image at the right time and the right position.

Reference: https://www.mindvision.ltd/

High-Speed Motion You Can't See? Capture the Critical Moment with the Right Resolution

calliope — Mon, 26 Oct 2026 07:57:56 +0000

The MindVision MV-XGC51GC/GM High-Speed Industrial Camera, Built for Motion Analysis and Equipment Debugging

Published October 22, 2026 · By MindVision Technology Co., Ltd. · For automation equipment manufacturers, pneumatic test equipment builders, vibratory feeder and parts-feeding suppliers, packaging machinery companies, and R&D and commissioning teams

High-Speed Imaging · Motion Analysis · Equipment Debugging · MindVision

Table of Contents
The Problem: When Motion Is Too Fast for Ordinary Video

Who Needs High-Speed Imaging — Not Just High-End Inspection Lines

Product Introduction: The MV-XGC51GC/GM High-Speed Industrial Camera

Key Capabilities: 10GigE Interface and Hardware Trigger

Application Scenarios: Observing High-Speed Motion on Four Machine Types

Selection Perspective: Don't Chase Megapixels Blindly

Beyond the Camera: A High-Speed Imaging System Is a Whole Chain

Frequently Asked Questions (FAQ)

Contact MindVision

The Problem: When Motion Is Too Fast for Ordinary Video
If you develop or commission automation equipment, these situations will likely feel familiar:

A pneumatic cylinder extends and retracts in tens of milliseconds. Ordinary video captures one or two frames of it — which instant the collision happens in, or where in the stroke the jamming occurs, is simply invisible.

Parts jumping inside a vibratory bowl feeder become a blurred mass in normal footage — how the part jammed or nested, no amount of replaying will show it.

Whether a feeding mechanism starts and stops smoothly, whether two mechanisms are synchronized — with eyesight and ordinary video you can only "feel something is off," without any evidence to locate the problem.

The common thread: it's not that the machine can't be filmed — it's that each frame is too far apart in time, so the critical part of the process gets skipped. Ordinary 25/30 fps video has 30–40 milliseconds between frames — and the critical phases of many mechanical actions complete within exactly those tens of milliseconds.

A different way to think about it: for this class of application, what you need may not be a higher-resolution camera, but one with a short enough frame interval — breaking the key motion into a dense sequence of image frames, so every frame becomes analyzable evidence.

Who Needs High-Speed Imaging — Not Just High-End Inspection Lines
When people hear "high-speed camera," they often think of scientific research or high-end production inspection lines. In reality, a significant share of high-speed imaging needs come from equipment manufacturers' own R&D and commissioning work:

Automation equipment manufacturers — machine development, mechanism motion verification, prototype commissioning: seeing high-speed motion clearly to analyze jamming, collision, and handoff problems.

Pneumatic component test equipment builders — cylinder stroke, pneumatic mechanism motion, and repeated-action testing: recording the critical phases of fast actions to analyze motion behavior and timing.

Vibratory bowl feeder and automatic feeding equipment suppliers — part vibration, orientation, transport, and feeding analysis: observing part motion states to troubleshoot feeding jams, part collisions, and mechanism anomalies.

Packaging machinery companies — mechanism start/stop, material transport, and motion handoffs: recording high-speed actions as reviewable image material for commissioning and fault analysis.

For these customers, the value of high-speed imaging is not "a beautiful high-resolution photo" but image-based evidence for analysis: in which frame did the problem occur, what state was the mechanism in, how large was the timing deviation — that is the information R&D and commissioning engineers actually need.

Product Introduction: The MV-XGC51GC/GM High-Speed Industrial Camera
For the scenarios above, MindVision offers the MV-XGC51GC/GM series (GC = color version, GM = monochrome version), positioned as a cost-effective high-speed industrial camera for motion analysis and equipment debugging.

Key Specification Value
Model MV-XGC51GC (color) / MV-XGC51GM (monochrome)
Maximum frame rate 1594.75 fps
Resolution at max frame rate 816 × 624
Data interface 10GigE
Sensor Sony IMX426
Pixel size 9.0 μm
Trigger Hardware trigger supported
Note the frame rate–resolution pairing: the MV-XGC51GC/GM's maximum frame rate of 1594.75 fps is achieved at a resolution of 816 × 624. Frame rate and resolution are always presented together — when evaluating, ask "is this frame rate enough?" and "is this resolution enough to see my target?", never just the frame rate number.

The large 9.0 μm pixels give each photosite a physically larger light-collecting area. The color GC suits applications where targets must be distinguished by color; the monochrome GM offers higher sensitivity for observation tasks that demand low-light detail. Both versions share the same platform, so R&D teams can choose flexibly per application.

Key Capabilities: 10GigE Interface and Hardware Trigger
High-speed imaging is not only about "a fast sensor." Two engineering capabilities determine how useful the camera is on the machine floor:

① 10GigE interface: the data channel for high-frame-rate imaging

Continuously capturing at 1594.75 fps produces a data stream far beyond what Gigabit Ethernet can carry. The 10GigE interface provides the matching bandwidth foundation, keeping high-frame-rate image data flowing steadily to the host. Ethernet architecture also supports longer cable runs — the camera does not have to sit next to the PC — which for compact automation equipment means far more flexible mounting positions on site.

② Hardware trigger: precise synchronization with machine actions

When observing mechanism motion, "when recording starts" matters as much as "what is recorded." The MV-XGC51GC/GM supports hardware trigger and can receive external signals from the machine's control system — a solenoid valve energizing, a feeding mechanism starting, a button press — starting acquisition at the very moment the action occurs. Every clip then has an explicit time anchor, giving multi-sequence comparison and cross-mechanism synchronization analysis a reliable basis.

Why it matters: bandwidth answers "can it be captured and transmitted"; triggering answers "is it captured at the right moment." Together, they turn a high-speed camera from "a device that shoots slow motion" into "a measurement tool that supports motion analysis."

Application Scenarios: Observing High-Speed Motion on Four Machine Types
The following describes the intended application directions of the MV-XGC51GC/GM — what high-speed imaging can offer in these scenarios. These are not verified customer case studies, and individual results depend on the complete setup.

Scenario 1: Automation equipment — mechanism verification and prototype commissioning
Does a new prototype's pick-and-place or transfer mechanism jitter, collide, or hand off unsmoothly at full speed? High-speed continuous capture breaks one complete action into a dense frame sequence; engineers step through frame by frame to locate exactly when a jam or collision occurs and in what mechanism state — evidence for design changes.

Scenario 2: Pneumatic component testing — cylinder stroke and repeated-action analysis
Cylinder extension and retraction is a classic millisecond-scale action. High-speed imaging records the critical phases of the full stroke; combined with hardware triggering aligned to the drive signal, it supports analysis of motion smoothness and timing consistency across repeated cycles.

Scenario 3: Vibratory bowl feeders and feeding equipment — part motion observation
Parts jumping, flipping, and orienting inside a bowl are nearly indecipherable in ordinary video. High-speed capture reveals the true motion state of parts on the track, helping localize where feeding jams, part nesting, or track blockages occur and why.

Scenario 4: Packaging machinery — start/stop handoffs and material transport
Mechanism start/stop and material cutting-and-handoff are focal points of packaging machine commissioning. High-speed footage records these transient processes as reviewable material, letting commissioning engineers compare mechanism behavior under different parameter settings.

To be clear: a high-speed camera provides image-based evidence. It helps engineers locate problems faster and describe them more precisely — but solving the problem itself still depends on the engineer's own design and adjustments.

Selection Perspective: Don't Chase Megapixels Blindly
For "watching mechanical motion" applications, a common selection mistake is to assume higher resolution is always better. In fact, resolution, frame rate, and data bandwidth constrain one another:

Higher resolution means more data per frame, which typically limits the achievable frame rate.

For motion observation, "frames per second" often determines whether you can see the critical process at all — more so than "pixels per frame."

An 816 × 624 resolution window is usually sufficient to show the motion detail needed when observing medium-field-of-view actions such as cylinder strokes, part orientation, or mechanism start/stop.

The sounder approach is to work backward from the application:

Motion speed: how fast is the action? What frame interval (in milliseconds) does it require?

Field of view: the whole mechanism, or a local section of the stroke?

Target size: how large is the feature of interest (a part, a cylinder rod end) within the field of view?

Required detail: do you need to resolve pose changes and position deviations, or just see the action sequence?

When the answers point to "medium field of view, very fast action, seeing the process matters more than seeing surface texture," a camera like the MV-XGC51GC/GM — frame-rate first, resolution sufficient — often fits the need better than a high-megapixel option, at a better cost point.

Beyond the Camera: A High-Speed Imaging System Is a Whole Chain
One final point: the high-speed camera is the core component of a high-speed imaging system, but not the whole system. Real-world results are also influenced by:

Lighting: at high frame rates the per-frame exposure window is very short; sufficiently bright illumination is usually needed to avoid underexposed frames.

Exposure time: shorter exposure reduces motion blur but raises lighting requirements — a balance to strike.

Lens: image quality, sensor-format coverage, and mounting space all affect the final result.

Host PC performance: reception and processing capability must match the data rate.

Data storage: sustained high-speed capture generates large volumes; storage and write throughput need advance planning.

Software: acquisition stability and playback/analysis capability directly shape the day-to-day experience.

The bottom line: when high-speed motion is invisible, the problem is usually not "not enough pixels" but "not enough frames per second." Choose the right resolution–frame-rate combination, back it with hardware triggering and adequate system support, and break the critical moment into analyzable frames — that is the real value a high-speed industrial camera brings to equipment R&D and commissioning.

If your team is looking for a high-speed imaging solution for motion analysis, mechanism commissioning, or fault tracing, get in touch with MindVision. Tell us your motion speed, field-of-view needs, and observation targets — our engineers will help confirm whether the MV-XGC51GC/GM configuration matches your application.

Frequently Asked Questions (FAQ)
Q1: What is the maximum frame rate of the MV-XGC51GC/GM, and at what resolution?
The maximum frame rate is 1594.75 fps, achieved at a resolution of 816 × 624. Frame rate and resolution are always bound together — higher frame rates generally correspond to smaller resolution windows — so the two figures should always be evaluated together rather than looking at the frame rate number alone.

Q2: Can a high-speed industrial camera directly solve machine jamming, collision, and synchronization problems?
Not guaranteed. A high-speed camera's role is to record dense motion sequences and provide frame-by-frame image evidence, helping engineers pinpoint when and in which state a problem occurs — which improves the efficiency of analysis and debugging. Solving the underlying problem still depends on the engineer's design and adjustments to the machine.

Q3: Do I need the highest possible resolution when choosing a high-speed camera?
No. Resolution, frame rate, and data bandwidth constrain each other: higher resolution means more data per frame, which typically limits the achievable frame rate. For observing mechanical motion, 816 × 624 resolution combined with over 1500 fps often captures more of the critical motion detail than a high-resolution, low-frame-rate alternative. Selection should be based on motion speed, field of view, target size, and the level of detail the application actually requires.

Q4: What does a 10GigE interface contribute to high-speed imaging?
A 10GigE interface provides far more bandwidth than Gigabit Ethernet — the channel foundation for continuously transmitting high-frame-rate image data. Ethernet cabling also supports longer deployment distances, giving flexible installation options on the machine floor. Actual throughput also depends on the host NIC, link quality, and software processing capability.

Q5: Besides the camera, what else does a high-speed imaging system require?
At high frame rates the exposure window per frame is very short, so sufficiently bright lighting is usually required. You should also consider lens quality and sensor-format matching, host PC reception and processing performance, data storage and write speed, and the acquisition software's stability and playback/analysis capability. The camera is the core component, but overall performance is determined by the entire chain.

Q6: What applications is the MV-XGC51GC/GM suited for?
It is aimed at high-speed motion observation and equipment debugging: mechanism motion verification and prototype debugging on automation equipment, pneumatic cylinder stroke and repeated-action testing, part motion analysis in vibratory bowl feeders and feeding systems, and start/stop and handoff observation on packaging machinery. The color MV-XGC51GC suits applications where color distinction matters; the monochrome MV-XGC51GM offers higher sensitivity for low-light detail.

Reference: https://www.mindvision.ltd/

3D Cameras vs. 2D Cameras: Core Differences, Selection Logic, and 3C Precision Assembly Inspection Applications

calliope — Wed, 21 Oct 2026 08:38:06 +0000

I. Introduction: Why Is 3D Vision Entering Precision Assembly Inspection?
In industrial automation and machine vision, 2D industrial cameras have long been used for appearance inspection, character recognition, dimensional measurement, and defect screening. However, when inspection tasks extend from "understanding flat images" to "measuring height, identifying deformation, analyzing 3D profiles, and locating spatial pose," the data-dimension boundaries of 2D vision become apparent.

For example, in 3C consumer electronics manufacturing, precision structural components such as mobile phone middle frames, connectors, camera modules, and shielding covers often require inspection of assembly height, local step differences, flatness, warpage, and dents. A 2D camera can detect some anomalies through grayscale, edge, and texture variations, but it is difficult for it to directly provide true Z-axis height and geometric deviation. A 3D camera can output depth maps or 3D point clouds, providing more direct geometric data for height measurement, profile analysis, and spatial positioning.

It should be emphasized that 3D vision is not a complete replacement for 2D vision, but a supplement and upgrade for requirements involving real spatial information.

II. Essential Differences Between 2D and 3D Cameras
2.1 Definitions: What Are 2D and 3D Cameras?
A 2D industrial camera mainly captures pixel intensity, color, and texture information on the imaging plane and outputs 2D images. It is suitable for character recognition, color inspection, printing inspection, 2D dimensional measurement, and surface-detail analysis.

A 3D industrial camera mainly captures depth, spatial coordinates, and geometric topography information of object surfaces, and outputs depth maps, point clouds, or 3D data. It is suitable for height measurement, step-difference inspection, flatness analysis, warpage and deformation inspection, 3D profile analysis, and robotic spatial positioning.

The core difference between the two is not merely resolution or image quality, but the way data is represented: a 2D camera records "how it looks," while a 3D camera further records "how it is shaped and positioned."

2.2 Comparison Between 2D and 3D Cameras
Comparison Dimension 2D Camera 3D Camera
Core data Pixel intensity, color, and texture Depth, spatial coordinates, and surface geometry
Spatial dimensions X and Y image coordinates X, Y, and depth-related Z information
Surface defect inspection Strong in color, texture, printing, and detail recognition Suitable for dents, protrusions, deformation, step differences, and profile inspection
Height measurement In many cases, additional equipment or assumptions are required Can directly support depth-based measurement
Lighting sensitivity Susceptible to lighting, shadows, and reflections Structured light relies on projected patterns but is still affected by optical conditions
Typical applications OCR, color inspection, printing inspection, and 2D dimensional measurement Height inspection, 3D profiling, deformation analysis, and spatial positioning
2.3 Limitations of 2D Vision
The main limitation of 2D vision lies in data dimensions, not in insufficient inspection capability.

It lacks direct depth information. A 2D image records the projection of an object on the imaging plane. For dents, protrusions, and height differences, additional imaging conditions, algorithmic inference, or measurement equipment are often required.

It is sensitive to illumination conditions. Light-source angle, brightness, exposure, and surface reflection can all change grayscale distribution, and the same defect may appear inconsistent under different lighting conditions.

Shadows and reflections can easily interfere with inspection. Polished metal, transparent materials, curved surfaces, and highly reflective parts may cause false judgments due to shadows or highlights.

Complex 3D topography is difficult to quantify directly. For dent depth, surface height differences, warpage, and 3D profiles, a 2D grayscale image alone generally cannot provide reliable true depth data.

2.4 Core Advantages of 3D Vision
The value of a 3D camera lies in capturing depth and shape information in real space, rather than generating an image that merely "looks more three-dimensional."

Direct access to depth data. It can output depth maps or point clouds, providing a foundation for Z-axis height measurement, profile analysis, and spatial positioning.

Better suited to complex surface topography inspection. For dents, protrusions, warpage, assembly height differences, and irregular profiles, judgments can be made from the perspective of geometric changes.

Reduced dependence on color and texture. When defects are primarily defined by height and shape, 3D vision can reduce reliance on color, printed patterns, and texture differences. However, highly reflective, transparent, dark, or low-reflectivity surfaces may still affect measurement quality.

Support for 3D dimensional analysis and automated positioning. Through point-cloud processing and measurement algorithms, height differences, surface profiles, and object pose can be analyzed, providing a basis for robotic picking, assembly, and quality control.

III. Why Evaluate the MindVision MV-3DF300G-901 3D Camera?
For companies seeking to introduce depth information into industrial inspection workflows, selecting a 3D camera should not rely on a single metric. It should be comprehensively evaluated based on actual workpieces, inspection tolerances, installation space, surface characteristics, and software integration capabilities.

The MindVision MV-3DF300G-901 is a 3D camera based on a single-camera structured-light principle with a DLP projection module. According to product documentation, the device supports both CPU and GPU processing modes and can output PLY-format point clouds, depth maps, and brightness maps. Its software supports secondary development, making it suitable for fields such as industrial manufacturing, inspection, consumer electronics, healthcare, and robotics.

3.1 Structured-Light Imaging and 3D Data Output
The MV-3DF300G-901 can output depth maps and PLY-format point clouds. Depth maps are suitable for observing depth variations at different surface positions, while point clouds can be used for 3D profile analysis, spatial coordinate calculation, and dimensional measurement. For vision systems that need to identify height differences, determine surface deformation, or extract spatial features of parts, these data can provide more direct geometric information for downstream algorithms.

In terms of key specifications, the model offers a resolution of 2048 × 1536, Z-axis accuracy of ≤ 0.06 mm, a recommended working distance of 500–1100 mm, a near-end field of view of 435 × 325 mm and a far-end field of view of 960 × 720 mm, and a typical acquisition time of approximately 1.6 s. These specifications can be used for an initial assessment of whether it is suitable for 3C structural components, assembly modules, and height/step-difference inspection within a medium field of view.

The device supports CPU and GPU processing modes. Product documentation indicates that GPU mode can provide higher processing efficiency, but in specific projects, overall performance should still be validated in combination with the point-cloud processing pipeline, computing platform, and acquisition cycle.

3.2 Industrial Integration Capabilities
The value of industrial vision equipment lies not only in data acquisition but also in whether it can be effectively integrated into the complete inspection workflow. The MV-3DF300G-901 supports a C++ SDK interface and secondary software development, providing an interface foundation for custom inspection logic, measurement application development, or integration with existing vision software. Its depth maps, brightness maps, and point clouds can be used in different analysis workflows according to application requirements.

During project evaluation, it is recommended to simultaneously validate data interfaces, algorithm compatibility, installation space, target surface characteristics, and the suitability of the device for the production cycle time.

3.3 Application Boundaries: 3D Cameras Are Not Universal Inspection Devices
3D cameras are suitable for inspection tasks defined by geometric shape, such as height, step difference, flatness, warpage, dents, and 3D profiles. However, for missing characters, color anomalies, printing defects, and extremely fine scratches, a 2D camera may still be more straightforward. In actual projects, combined 2D and 3D inspection is often more reasonable than a single technology.

IV. Segmented Application: Assembly Height and Deformation Inspection of Precision Structural Components in 3C Consumer Electronics
4.1 Typical Workpieces and Inspection Pain Points
Precision structural components in 3C consumer electronics manufacturing are usually small in size, have complex assembly relationships, require high appearance quality, and are subject to strict production cycle times. Typical inspection objects include mobile phone middle frames, camera module brackets, connectors, shielding covers, battery covers, and precision metal parts.

Common inspection requirements on actual production lines include:

Local step differences and height differences after assembling the mobile phone middle frame and shielding cover;

Assembly height and flatness of camera module brackets;

Coplanarity and local floating height of connector terminals;

Warpage, dents, and local deformation of shielding covers and battery covers;

Screw floating height and whether spring contacts are fully seated;

Robotic picking and positioning assistance for randomly oriented incoming parts or stacked parts.

The common point of these tasks is that defects or assembly states are reflected not only in color and texture, but more importantly in true height, flatness, and 3D profile. A 2D image can show edges and brightness variations, but it is difficult for it to directly answer "how much higher," "how much difference," or "whether it is warped."

4.2 Boundaries of 2D Vision in 3C Precision Assembly
A 2D camera is suitable for character recognition, color inspection, printing inspection, and 2D dimensional measurement. However, 2D vision has clear boundaries in the following tasks:

The task requires determining true Z-axis height or assembly step difference;

Defects mainly appear as dents, protrusions, or local deformation;

Metal surface reflections cause unstable grayscale features;

Parts are stacked, change pose, or have height differences;

Measurement results need to be compared with 3D CAD models or reference profiles.

For example, in mobile phone middle-frame or shielding-cover assembly inspection, a 2D image may show brightness variations but can hardly directly determine local dent depth or assembly surface height difference. In this case, 3D depth data is more suitable for geometric deviation judgment.

4.3 Typical 3D Camera-Based Inspection Workflow
For 3C precision structural components, actual projects can adopt the following workflow:

Establish a reference plane or reference model. Extract assembly surfaces, reference planes, or standard profiles from acceptable samples.

Acquire depth maps and point clouds. Use a 3D camera to obtain surface geometry data of the part under inspection.

ROI segmentation and key-region extraction. Set inspection regions for assembly surfaces, connectors, camera-module areas, and other key regions.

Calculate height, step difference, flatness, and warpage. Calculate local height deviation, regional height difference, and flatness metrics based on point clouds or depth maps.

Tolerance judgment and result output. Compare measured values with predefined tolerances, output OK/NG results, and connect to PLC, MES, or robot control systems.

Combined analysis with 2D data. For characters, colors, printing, and fine scratches, a 2D camera can still be retained for complementary inspection.

4.4 Practical Value and Limitations of 3D Vision in This Segment
In 3C precision assembly inspection, the core value of 3D vision is to provide quantifiable geometric information, not merely image enhancement. For problems defined by shape, such as assembly height, step difference, flatness, warpage, and dents, depth maps and point clouds can help the system make judgments from a geometric perspective and reduce reliance on color and texture features.

However, practical implementation still requires attention: highly reflective metal, mirror surfaces, transparent adhesive, and dark low-reflectivity surfaces may affect structured-light measurement quality; production-line vibration, ambient light, and temperature changes may also affect repeatability; point-cloud processing algorithms, calibration methods, and cycle-time matching need to be validated together with camera selection. Therefore, in the early stage of a project, real workpieces should be used for imaging tests and parameter optimization, rather than judging suitability solely based on sample images or a single metric.

4.5 Match with the MV-3DF300G-901
The depth maps, brightness maps, and PLY point-cloud outputs of the MV-3DF300G-901 can be used for height, step-difference, flatness, and profile analysis of 3C structural components. Its C++ SDK and secondary-development capabilities can provide a foundation for customized vision system integration. With its 2048 × 1536 resolution, ≤ 0.06 mm Z-axis accuracy, 500–1100 mm recommended working distance, and approximately 1.6 s typical acquisition time, it can be evaluated as a candidate solution for assembly component inspection within a medium field of view. However, final suitability still needs to be validated through sample testing based on the surface reflections, assembly tolerances, installation space, and production cycle time of real workpieces.

V. How to Choose Between 2D and 3D?
3D vision is not a complete replacement for 2D vision. The correct selection approach is to determine what data is needed based on the inspection objective.

Inspection Requirement Recommended Approach Reason
Color identification 2D The core requirement is color information
OCR and character inspection 2D Better suited to character and image-detail analysis
Fine surface scratches Usually prioritize 2D; evaluate 3D if measurable topographic changes are formed Depends on scratch depth, width, and contrast
Dent depth measurement 3D Requires depth information
Component height measurement 3D Supports Z-axis measurement
Warpage and deformation 3D Enables geometric profile analysis
Robotic spatial positioning Usually prioritize 3D evaluation when height and pose are involved Provides spatial geometry data
Combined color and depth inspection Combination of 2D and 3D Uses complementary visual information
VI. Frequently Asked Questions (FAQ)
Q1: Is a 3D camera always better than a 2D camera?
Not necessarily. A 3D camera excels at depth, shape, and spatial information. If the task mainly involves color, characters, print quality, or fine texture, a 2D camera is often more straightforward and cost-effective. For height measurement, dent inspection, flatness inspection, and 3D positioning, 3D vision should be evaluated first. Whether a 3D industrial camera is suitable should be judged based on the inspection objective, rather than simply comparing camera grades.

Q2: Can a 3D camera detect surface defects that a 2D camera cannot?
In some applications, yes, especially when defects mainly appear as dents, protrusions, step differences, or height variations. However, fine scratches, color anomalies, and printing defects may still require 2D imaging. For surface defect inspection of 3C structural components, a more reasonable approach is to first determine whether the defect creates measurable geometric changes, and then decide whether to use 2D, 3D, or combined 2D and 3D inspection.

Q3: Can a 3D camera completely eliminate the effects of lighting and reflections?
No. A structured-light 3D camera acquires depth information by projecting and analyzing optical patterns, which can reduce some dependence on natural brightness and shading cues. However, highly reflective, transparent, dark, or low-reflectivity materials may still affect measurement results. For 3D inspection of reflective metal, mirror structures, and transparent adhesive areas, validation should be carried out through real-sample testing, exposure parameter optimization, and algorithmic processing.

Q4: What data outputs and integration methods does the MV-3DF300G-901 support?
According to product documentation, the device can output brightness maps, depth maps, and PLY-format point clouds, and supports CPU/GPU processing modes, a C++ SDK interface, and secondary software development. It offers a resolution of 2048 × 1536, Z-axis accuracy of ≤ 0.06 mm, a recommended working distance of 500–1100 mm, and a typical acquisition time of approximately 1.6 s. Specific compatibility still needs to be confirmed based on the operating system, software architecture, data-processing workflow, and project requirements. For projects requiring C++ SDK machine vision integration or PLY point-cloud inspection, interface and algorithm integration methods can be key validation points during evaluation.

Q5: Is a 3D camera suitable for step-difference and flatness inspection of mobile phone middle frames?
Suitable for evaluation. Middle-frame step differences, flatness, and local deformation are usually defined by true height variations, which are difficult for 2D images to quantify directly. A 3D camera can establish a reference plane through depth maps or point clouds and calculate local height deviation, regional step difference, and flatness metrics. Practical implementation requires attention to middle-frame surface reflections, reference-plane selection, fixturing stability, and production cycle time. This type of application can be categorized as mobile phone middle-frame step-difference inspection, middle-frame flatness inspection, and 3D vision inspection of 3C precision structural components.

Q6: Should shielding-cover assembly height inspection use 2D or 3D?
If the inspection objective is whether the shielding cover is fully seated, local floating height, assembly height difference, or coplanarity, 3D vision should usually be evaluated first. A 2D camera can observe edges and brightness variations but can hardly provide true height directly. Shielding-cover assembly height inspection is better suited to using depth maps or point clouds for reference-plane comparison and height-deviation calculation. If character, color, or printing inspection is also required, a 2D camera can be retained for combined inspection.

Q7: How can 3D inspection be used for connector coplanarity and terminal floating height?
Connector coplanarity and terminal floating height are typical 3D geometric inspection tasks. A 3D camera can acquire point clouds of the terminal area, extract key-point heights, calculate coplanarity, local floating height, and height differences, and compare them with the tolerance range. This type of application can be searched as connector coplanarity inspection, terminal floating-height inspection, and connector 3D vision measurement. Actual projects need to consider small terminal size, strong reflections, and point-cloud resolution matching.

Q8: How can camera-module assembly height inspection be implemented with a 3D camera?
Camera-module assembly height inspection usually needs to determine height differences and assembly flatness among the module, bracket, or surrounding structural parts. A 3D camera can acquire depth data of the assembly area, establish a reference plane, extract module-area height, and calculate deviation. Camera-module assembly height inspection is suitable for combined use with a 2D camera: 3D handles height and step difference, while 2D handles characters, appearance, and surface details.

Q9: What should be considered when using structured-light 3D inspection for reflective metal 3C structural components?
Highly reflective metal, mirror surfaces, dark low-reflectivity surfaces, and transparent materials may affect structured-light measurement quality. During project evaluation, attention should be paid to whether surface reflections cause missing point clouds; whether multiple captures are stable; whether ambient light and production-line vibration affect repeatability; and whether the algorithm can handle noise points. For 3D inspection of reflective metal, real-workpiece testing is more important than sample images.

Q10: Can a 3D camera replace 2D for character, scratch, and printing inspection?
Complete replacement is generally not recommended. Character recognition, color inspection, printing inspection, and extremely fine scratches are still better suited to 2D vision. 3D cameras have advantages in dent, protrusion, step-difference, warpage, and 3D profile inspection. Actual production lines are better suited to combined 2D and 3D vision inspection: 2D for color and texture, 3D for height and shape.

Q11: Why should 3D vision projects use real workpieces for sample validation?
The measurement quality of 3D vision is closely related to workpiece surface material, reflection characteristics, installation distance, ambient light, and algorithm processing. It is difficult to judge actual production-line stability solely based on product documentation or standard samples. 3D vision sample validation should cover real workpieces, actual installation pose, production cycle time, and tolerance-judgment logic, with emphasis on confirming point-cloud completeness, repeatability, measurement deviation, and algorithm stability.

Q12: What supporting components are required for integrating a 3D camera with robotic picking?
A 3D camera can provide point clouds and spatial geometry data for robotic bin picking, loading/unloading, and assembly positioning. However, a complete picking system also requires hand-eye calibration, coordinate transformation, object recognition, grasp planning, collision detection, and a robot control interface. 3D vision projects for robotic bin picking should not evaluate only the camera; cycle time, point-cloud processing time, and picking success rate must also be validated.

Q13: How can the suitability of a 3D camera under production cycle time be evaluated?
Do not look only at single acquisition time; acquisition, point-cloud processing, algorithm judgment, communication, and robot motion should be evaluated together. 3D vision production-cycle evaluation should use real workpieces and the actual computing platform to confirm whether the total time from trigger image acquisition to OK/NG result output meets production-line requirements. CPU/GPU processing modes, algorithm complexity, and data volume all affect the final cycle time.

VII. Conclusion: Choose Vision Technology Based on the Inspection Task
There is no absolute superiority or inferiority between 2D and 3D cameras; they address different data-acquisition and inspection problems. If the inspection objective is color, characters, print quality, or surface texture, 2D vision is usually more efficient and straightforward. If the task involves true height, dent depth, deformation, 3D profile, or spatial positioning, a 3D camera can provide more suitable geometric information.

In the inspection of 3C consumer electronics precision structural components, 3D vision is particularly suitable for inspection tasks defined by geometric shape, such as assembly height, step difference, flatness, warpage, and deformation. The MindVision MV-3DF300G-901 adopts a single-camera structured-light and DLP projection solution, supports depth maps, brightness maps, and PLY point-cloud outputs, and features CPU/GPU processing modes and a C++ SDK interface. Its 2048 × 1536 resolution, ≤ 0.06 mm Z-axis accuracy, 500–1100 mm recommended working distance, and approximately 1.6 s typical acquisition time can serve as initial evaluation criteria for medium-field-of-view 3C assembly inspection projects.

For manufacturers upgrading their machine vision systems, it is recommended to start from actual defect types, measurement tolerances, installation space, surface characteristics, and production cycle time, conduct sample testing first, and then determine the suitable camera and system configuration.

Reference: https://www.mindvision.ltd/

Connector Vision Inspection System: How Machine Vision Inspects Connectors and Terminals

calliope — Sat, 10 Oct 2026 02:47:03 +0000

A Complete Guide to Industrial Camera Selection for Connector and Terminal Inspection

Introduction: Connector Vision Inspection System Connectors and terminals are among the smallest but most critical components in modern electronic and electrical products. Although a connector may look simple from the outside, its internal terminals, pins, contact surfaces, and plastic housing often require extremely precise manufacturing. A bent pin, missing terminal, incorrect position, burr, crack, contamination, or dimensional deviation may affect assembly accuracy and electrical performance.

This is why automated visual inspection has become increasingly important in connector manufacturing. A modern machine vision for electronics inspection system can combine an industrial camera, lens, lighting, trigger, and image-processing software to automatically inspect connector appearance, terminal position, dimensions, and assembly quality.

Compared with manual inspection, machine vision provides a more repeatable imaging process and can be integrated directly into automated production lines.

Inspection Items and Typical Defects Connector inspection is not just about checking whether the product is dirty. In real production, a vision system may need to inspect appearance, dimensions, position, and completeness at the same time.

Common inspection items include terminal missing, bent pins, pin position, terminal spacing, housing defects, burrs, cracks, scratches, contamination, dimensions, hole diameter, assembly status, missing parts, and orientation.

Typical defects include missing, bent or deformed terminals; pin displacement or incorrect orientation; housing cracks, deformation or flash; surface scratches, stains or contamination; width, height, spacing or diameter deviation; missing or incorrectly assembled components; terminal offset or misalignment; and color or surface abnormalities.

Inspection Item Typical Defects
Terminal inspection Missing, bent or deformed terminals
Pin inspection Pin displacement or incorrect orientation
Housing inspection Cracks, deformation, flash
Surface inspection Scratches, stains, contamination
Dimensional inspection Width, height, spacing, diameter
Assembly inspection Missing or incorrectly assembled parts
Position inspection Terminal offset or misalignment
Appearance inspection Color and surface abnormalities

Why Connector and Terminal Inspection Is Difficult For automotive connectors, board-end connectors, and high-density electronic connectors, one product may have many very tiny pins, so the camera must not only capture the product but also see product details clearly.

One of the biggest challenges in connector vision inspection is that the target is small, but the inspection requirements are high. For example, a connector may be only 20–50 mm overall, while the pin pitch, terminal edge, or burr to be inspected may be only 0.1 mm or even smaller.

This means the industrial camera needs sufficient spatial resolution. For this type of application, an area scan camera for industrial inspection is a common choice, because it can capture a two-dimensional image of the entire connector area at one time and is suitable for connector appearance, terminal position, pin, housing, and dimensional inspection.

If the target moves at high speed on the production line, shutter type and exposure time must also be considered. In a machine vision camera industrial project, resolution, speed, and imaging stability must be evaluated together.

Resolution, FOV and Camera Selection Many people ask directly during industrial camera selection, “Is 5MP enough?” In fact, knowing only the camera pixel count is not enough to judge whether the inspection requirement can be met. You need to consider FOV + Sensor Resolution + Minimum Defect Size.

For example, assume connector width is 80 mm, FOV is 100 mm, and minimum defect is 0.1 mm. If one defect should cover about 3 pixels, then 0.1 ÷ 3 ≈ 0.033 mm/pixel, and 100 ÷ 0.033 ≈ 3030 pixels. Therefore, the horizontal resolution should be at least about 3000 pixels. If dimensional measurement or smaller defect detection is required later, some margin should be reserved.

This is why a 5MP camera for machine vision can handle some ordinary inspection tasks, while finer pin inspection may need a higher-resolution sensor. For large-FOV, high-precision inspection, a 25MP camera for high-resolution inspection can also be considered.

A 5MP sensor is suitable for ordinary connector appearance, larger terminals, general dimensional inspection, and medium-precision defect detection; a 12MP–20MP sensor is suitable for small connectors, dense pins, fine surface inspection, and large-FOV high-precision inspection; a 25MP sensor is suitable for large-FOV precision inspection, multi-region inspection, and extremely small defect detection.

Therefore, when choosing a sensor, do not simply pursue the highest pixel count; calculate based on the actual FOV. A machine vision camera buying guide should not simply rank cameras by MP count; it should start from the entire imaging system.

Sensor and Sony Options One of the cores of an industrial camera is the sensor. In machine vision projects, the sensor can be selected by resolution, shutter type, pixel size, frame rate, and sensitivity.

For some projects, engineers may compare a Sony IMX265 camera for machine vision with a Sony IMX264 5MP camera. The final choice needs to combine resolution, pixel size, frame rate, shutter type, exposure time, FOV, and lens resolution.

Therefore, a machine vision camera buying guide cannot simply rank by MP count; it must consider the complete imaging system.

Global Shutter for Moving Terminals Connector production lines usually move continuously. If the product shifts significantly during exposure, motion blur or geometric distortion may occur. This is especially important when inspecting pin straightness, terminal position, terminal spacing, and connector dimensions, because image distortion can directly affect algorithm judgment.

Therefore, for high-speed connector inspection, a high-speed machine vision camera + Global Shutter can be considered. A global shutter allows the entire sensor to capture the image within a unified exposure time, reducing geometric distortion caused by moving targets.

If inspection speed increases further, exposure time + frame rate + trigger frequency must also be considered.

Lens, FOV and Magnification No matter how good the sensor is, if the lens resolution is insufficient, ideal inspection results are difficult to obtain. For a fixed inspection position, a fixed focal length lens for machine vision can be used; it has a simple structure and is suitable for automated inspection equipment with a fixed FOV.

For small terminals and pins, a macro lens can be used. For precision dimensional measurement, a telecentric lens can be considered. Industrial vision systems also often use a C-mount camera lens for industrial use, because C-Mount has a mature industrial vision ecosystem and can be combined with different sensor sizes and industrial cameras.

Before selecting a lens, first determine the product size, the required viewing area, and the minimum defect size. For example, the product is 30 × 20 mm, and the whole product should be in the image while detecting a 0.05 mm pin defect. In this case, FOV, magnification, working distance, sensor size, and pixel resolution need to be calculated.

Engineers can use a field of view calculator for industrial camera to assist the calculation. However, theoretical calculations still need to be verified by real product tests, because lens resolution, optical distortion, and working distance all affect the final result.

Lighting, Polarization and ROI Lighting is critical for terminal inspection. Connector inspection often involves two completely different materials: plastic + metal. Plastic parts are easily affected by color, texture, and surface reflection, while the biggest difficulty of metal terminals is reflection.

If the lighting angle is unreasonable, metal pins may produce very strong highlights, causing overexposure in some areas, disappearing pin edges, unstable metal surface texture, and uneven brightness across the same product. Different lighting schemes can therefore be used to improve image quality.

For metal terminals with obvious reflection, a polarizing filter for machine vision can be considered. A typical optical path is LED Light → Polarizer → Connector → Camera. By adjusting the light source, polarization direction, and camera angle, some surface defects can be separated from the background more easily.

However, polarization is not required in every connector inspection; it still needs to be tested according to metal material, surface finish, lighting angle, and defect type.

If only a small area in a connector image needs to be inspected, there is no need to let the algorithm process the entire image. For example, the whole image may be 4000 × 3000 pixels, but only 1000 × 800 pixels is the terminal area that truly needs inspection. In this case, a camera ROI region of interest setup can be used.

ROI can reduce invalid data, improve processing efficiency, lower algorithm computation, increase effective frame rate in some applications, and facilitate different algorithms for different inspection regions.

Interfaces, Multi-Camera, PoE and Software Connector vision inspection often requires choosing between GigE and USB3. GigE Vision offers longer transmission distance, suitability for multi-camera applications, Ethernet networking, good high-speed transmission, flexible industrial integration, and multi-station support. USB3 Vision offers shorter distance, possible multi-camera use, USB structure, higher transmission speed, simple integration, and device-dependent multi-station capability.

For actual selection, refer to a GigE Vision vs USB3 Vision camera comparison. If the equipment contains multiple inspection stations, consider how to connect multiple GigE cameras.

Multi-camera connector inspection is well suited to GigE. For example, a connector may need inspection from four directions: Top View, Front View, Side View, and Terminal View. This is a typical multi-camera inspection system. Multiple industrial camera GigE units can be connected to a vision control computer through an industrial network.

One advantage of GigE is flexible wiring, and it is suitable for large automation equipment. If higher transmission bandwidth is needed, a 10GigE camera for high-resolution inspection can be considered; it is especially suitable for high-resolution sensors and high-data-volume image transmission.

PoE can simplify industrial camera installation by reducing the number of cables. A PoE GigE camera setup guide can be used to provide data transmission and power to the camera through PoE network equipment. Whether to use PoE should be designed according to camera power requirements, network switch, cable length, and number of cameras.

Industrial cameras also need stable communication with vision software. During project selection, pay attention to GenICam compatible camera selection. The GenICam ecosystem helps industrial cameras integrate with different vision software and development environments.

Some connector inspection projects may use HALCON, OpenCV, custom vision software, C++, or Python. For systems using HALCON, consider a machine vision camera compatible with Halcon. For companies using OpenCV for vision algorithm development, pay attention to OpenCV camera integration for industrial vision.

Color or monochrome depends on the inspection target. If the main inspection items are pin position, shape, dimension, edge, burr, and crack, a monochrome camera usually provides better grayscale contrast. If color difference, wrong color, color marking, or housing color must be identified, a color industrial camera is more valuable.

Use monochrome vs color camera for inspection to determine the sensor type. The core principle is that if color is not an inspection variable, there is no need to increase imaging complexity for color. This is also why how to choose machine vision camera interface should start from the whole equipment requirement, not only interface speed.

Applications, MindVision Solutions, FAQ and Conclusion Applications Automotive electronics is one of the important applications of connector vision inspection. An automotive connector may contain a plastic housing, multiple terminals, a locking structure, contact pins, and sealing components. Therefore, machine vision for automotive inspection can cover multiple inspection steps, such as checking whether terminals are missing, whether pins are bent, whether terminal positions are offset, whether the plastic housing has burrs, whether the locking structure is correctly assembled, and measuring key dimensions.

If the production line speed is high, high-resolution + Global Shutter + Short Exposure can be used, with suitable lens and lighting configured according to product size.

Consumer electronic connectors usually have small size, high pin density, and tight tolerance. Examples include board-to-board connectors, FPC connectors, USB connectors, SIM connectors, and micro connectors. For such small products, a high-resolution camera with a macro or high-resolution lens can be used.

A typical process is Feeding → Positioning → Image Capture → Terminal Inspection → Dimension Inspection → Defect Classification → OK/NG Sorting. The whole process can be integrated with PLC, robots, and automatic sorting mechanisms.

To design a complete connector vision inspection system, first define the defect, such as 0.05 mm, 0.1 mm, or 0.2 mm. Then define the FOV, calculate resolution by FOV ÷ Pixel Number, select the camera, select the lens, design lighting, and select the interface.

Common problems include unclear pin edges, overexposed metal terminals, distorted moving terminals, unstable multi-camera systems, and clear images that still fail in the algorithm. A good image does not equal a good inspection result. Therefore, a vision inspection system must be optimized from optics, hardware, and software together.

MindVision Solutions
For small precision parts such as connectors and terminals, MindVision can provide industrial cameras according to different inspection needs. High-resolution industrial cameras are suitable for small connector inspection, fine terminal inspection, pin inspection, surface defect inspection, and dimensional inspection. GigE industrial cameras are suitable for multi-camera inspection, long-distance transmission, automated production lines, and large inspection machines. USB3.0 industrial cameras are suitable for high-speed image acquisition, single-camera inspection, and compact inspection equipment. Global Shutter cameras can reduce distortion caused by high-speed moving terminals and connectors. For OEM equipment, customized cameras can be provided for special needs in camera size, interface, sensor, housing, mounting, and cable direction.

Connector Inspection Camera Selection Table
Application Camera Direction Lens Direction Interface
Ordinary connector appearance 2–5MP area scan Fixed focal length lens GigE / USB3
Small connectors 5MP or above Macro/high-resolution lens GigE / USB3
Dense pins High-resolution Global Shutter High-resolution lens GigE / USB3
Precision dimensional measurement High-resolution camera Telecentric lens GigE
High-speed terminal inspection High-speed + Global Shutter High-resolution lens GigE / USB3
Multi-view inspection Multiple industrial cameras Selected by FOV GigE
Large-FOV high-precision inspection 12MP/20MP/25MP High-resolution lens 10GigE etc.
Reflective metal terminals Mono/Color as needed Polarization optics GigE / USB3
OEM equipment Customized According to structure According to project
FAQ
Q1. What camera is best for connector inspection?
There is no single industrial camera suitable for all connector inspection projects; choose according to product size, FOV, minimum defect, speed, precision, and interface.

Q2. Do connector inspections require high-resolution cameras?
For dense pins, small connectors, and tiny defects, high-resolution cameras are usually more suitable, but the final resolution requirement must be calculated with FOV.

Q3. Should I use a global shutter camera?
If the product moves at high speed during inspection, Global Shutter is usually more suitable for reducing image distortion.

Q4. Is GigE better than USB3 for connector inspection?
Both interfaces have their own suitable scenarios; for multi-camera, long-distance, and large inspection systems, GigE can be considered, while for short-distance, high-speed, and compact systems, USB3 Vision is also a common choice.

Q5. Do I need a telecentric lens?
For appearance defect inspection, a telecentric lens is not always necessary; for high-precision dimensional measurement such as pin spacing, terminal width, hole diameter, and connector dimensions, a telecentric lens can be considered.

Q6. How many cameras are needed for connector inspection?
It depends on the number of inspection surfaces; if only top appearance is inspected, one camera may be enough, but if Top + Front + Side + Terminal are inspected simultaneously, a multi-camera solution may be needed.

Conclusion
In conclusion, connectors and terminals are small, but their vision inspection involves a complete industrial imaging system. From defect definition to FOV → Resolution → Sensor → Lens → Lighting → Shutter → Interface → Trigger → Software, every step affects the final inspection result.

For connectors, terminals, automotive electronics, and consumer electronics parts, machine vision can detect missing, bent, burr, crack, and contamination defects, and can also complete position inspection, dimensional measurement, and assembly confirmation.

MindVision provides high-resolution industrial cameras, GigE industrial cameras, USB3.0 industrial cameras, Global Shutter cameras, line scan cameras, and OEM/ODM customization. For manufacturers developing connector inspection machines, terminal inspection systems, and automated connector inspection equipment, a suitable machine vision solution can be built from specific products and defect requirements.

Reference: https://www.mindvision.ltd/

From Mold to Inspection: How Vision Systems Check Plastic Parts

calliope — Sat, 10 Oct 2026 01:51:44 +0000

Plastic parts and injection-molded components are widely used in automotive, electronics, appliances, medical devices, packaging, and industrial equipment. As production speeds increase, manual visual inspection becomes increasingly difficult to maintain consistent quality. As a result, machine vision for plastic inspection is becoming an important part of automated quality control.

A complete plastic inspection system typically includes an industrial camera, lens, lighting, trigger, image-processing software, and automation equipment. The industrial camera converts the physical appearance of the product into a digital image that can be analyzed, making camera selection an important factor in the overall inspection performance.

For injection-molded parts, the challenge is not simply taking a picture. The real challenge is obtaining stable, clear, and repeatable images under high-speed production conditions, complex surfaces, and small defect sizes.

Why Is Machine Vision Important for Injection Molding Inspection?

During injection molding, variations in temperature, pressure, mold conditions, material flow, and process parameters can affect the final product quality.

As a result, multiple types of defects may appear within the same production batch, including:

Flash

Short shots

Sink marks

Cracks

Scratches

Bubbles

Deformation

Contamination

Color abnormalities

Dimensional deviations

Incorrect hole position

Assembly errors

Manual inspection relies heavily on operator experience, while machine vision can continuously operate according to predefined imaging and inspection criteria.

An industrial machine vision camera can therefore be used not only for appearance inspection but also for measurement, positioning, OCR, barcode recognition, and assembly inspection.

Common Visual Defects in Injection-Molded Parts

2.1 Flash Inspection

Flash commonly appears along mold parting lines or product edges.

For small plastic components, flash may only be tens or hundreds of micrometers in size, requiring sufficient spatial resolution.

This is a typical application for a camera for small parts inspection.

2.2 Short Shot Inspection

A short shot usually changes the expected contour of the molded component.

The industrial camera can capture the component contour, which can then be analyzed using edge detection, template matching, or other vision algorithms.

2.3 Sink Marks and Surface Depressions

Sink marks are common surface defects, but not every sink mark can be easily detected under conventional lighting.

For these applications, lighting direction, illumination angle, surface reflectivity, and lens resolving capability are all important.

2.4 Crack and Scratch Inspection

Cracks and scratches are often small defects with relatively low contrast.

Simply increasing the number of pixels does not necessarily solve the problem.

The system should also consider:

Camera resolution + lens resolution + lighting contrast + exposure time + image-processing algorithms

How to Select an Industrial Camera for Plastic Inspection?

Industrial camera selection should not begin with the question, “How many megapixels do I need?”

A more practical selection process is:

Inspection target → Minimum defect → Field of view → Pixel resolution → Production speed → Frame rate → Exposure time → Shutter → Lens → Lighting → Interface → System validation

This workflow can also serve as a practical industrial camera selection guide for beginners.

Step One: Determine Camera Resolution Based on the Minimum Defect

How to choose industrial camera resolution is one of the most important questions in an industrial vision project.

For example, assume the inspection field is 100 mm wide and the system needs to identify a defect approximately 0.1 mm in size.

If the defect should cover at least three pixels:

Pixel size ≈ 0.1 ÷ 3 = 0.033 mm/pixel

Therefore:

Required horizontal resolution ≈ 100 ÷ 0.033 ≈ 3000 pixels

A camera with more than approximately 3MP may therefore be considered, with additional margin depending on lens performance, product positioning tolerance, and algorithm requirements.

It is important to note that this is an engineering estimate rather than a guarantee of final inspection accuracy.

Actual system performance should be verified using representative samples.

High Resolution Does Not Automatically Mean High Inspection Accuracy

Many users looking for a high-resolution machine vision camera assume that more pixels always mean better inspection.

In practice, this is not necessarily true.

For example:

25MP camera + low-resolution lens

may perform worse than:

12MP camera + high-resolution lens + optimized lighting

The reason is that final image quality depends on multiple factors.

One important relationship is:

lens resolution vs sensor pixel size

If the sensor has very small pixels but the lens does not provide sufficient resolving power, the lens can become the bottleneck of the imaging system.

For high-resolution plastic inspection, the sensor, lens, and lighting should therefore be considered together rather than focusing only on megapixels.

Why Do Small Injection-Molded Parts Require High-Quality Imaging?

Connector housings, plastic gears, precision clips, and other small components are often compact while containing multiple fine structures.

This is a typical small parts inspection camera application.

The camera may need to resolve:

Product contour

Small holes

Edges

Clips

Grooves

Flash

Cracks

Markings

Therefore, sensor size and lens field of view must also be considered in addition to resolution.

Which Shutter Should Be Used for High-Speed Injection Molding Lines?

When plastic parts move rapidly on a conveyor, their position can change during exposure.

An unsuitable shutter configuration can result in motion blur or geometric distortion.

For high-speed vision inspection, a global shutter camera for motion capture is therefore worth considering.

Particularly for:

High-speed conveyors

High-speed sorting

Continuous production lines

High-speed positioning

Dynamic dimensional inspection

Global shutter technology can reduce the risk of motion-related geometric distortion associated with sequential pixel exposure.

Exposure Time and Frame Rate Also Matter

Even with a global shutter, an excessively long exposure time can still cause motion blur.

High-speed plastic inspection should therefore consider:

Object speed + exposure time + frame rate + lighting intensity

For example, stronger lighting can allow the exposure time to be reduced while maintaining sufficient image brightness.

Therefore, how to improve image quality from industrial camera is not simply a matter of adjusting camera settings. Lighting and exposure parameters are equally important.

What Lens Should Be Used for Plastic Inspection?

The lens is one of the most frequently overlooked components in an industrial vision system.

In practice, how to choose the right lens for industrial camera should consider:

Sensor size

Working distance

Field of view

Focal length

Magnification

Distortion

Optical resolution

9.1 Fixed-Focal-Length Lenses

For general plastic appearance inspection, a fixed focal length lens can be selected based on the field of view and working distance.

9.2 Macro Lenses

For small injection-molded components or fine defect inspection, a macro lens for industrial inspection can be considered.

It is suitable for applications requiring higher magnification.

9.3 Telecentric Lenses

When inspection progresses from simply detecting defects to measuring dimensions, the optical configuration needs to become more controlled.

For example:

Hole diameter measurement

Outer diameter measurement

Thickness measurement

Plastic gear dimensions

Precision molded-part profile

These applications can consider a telecentric lens for measurement applications.

GigE or USB3: Which Interface Should Be Used?

The interface should be selected according to the machine architecture rather than simply asking which interface is faster.

GigE Vision

GigE Vision cameras are well suited to production lines and multi-camera systems.

The keyword GigE Vision camera for multi-camera inspection represents a typical application scenario.

Key advantages include:

Network-based deployment

Long-distance transmission

Multi-camera expansion

Industrial automation integration

MindVision's product portfolio includes GigE area scan and line scan cameras as well as multiple network-interface options.

USB3 Vision

USB3 Vision is suitable for high-speed acquisition systems where the camera is relatively close to the host computer.

Relevant long-tail keywords include:

USB3 Vision camera for machine vision

industrial camera USB 3.0

USB3 Vision interface advantages

USB3 solutions can be used for:

Single-camera inspection

Compact vision equipment

Laboratory inspection equipment

High-speed short-distance acquisition

Multi-Camera Inspection for Injection-Molded Parts

For complex plastic components, one camera may not be able to cover every inspection area.

For example, an automotive plastic component may require:

Camera 1: Front-side appearance

Camera 2: Side flash inspection

Camera 3: Hole inspection

Camera 4: Character/OCR inspection

A system based on GigE Vision camera for multi-camera inspection can be considered for this type of architecture.

Multiple cameras can be connected to the host through an industrial network and positioned according to inspection stations and areas.

MindVision's GigE products support multi-camera operation and network-based deployment, according to its product documentation.

Why Are Triggering and Synchronization Important?

An industrial camera does not necessarily capture an image every time it sees a product.

In automated production lines, a common workflow is:

Photoelectric sensor detects product → Trigger signal → Camera exposure → Lighting synchronization → Image acquisition → Algorithm decision → PLC action

Therefore, a camera I/O trigger wiring guide can be an important technical resource for machine builders and engineers.

For high-speed inspection, synchronization between the camera, lighting, and motion system can directly affect inspection stability.

How to Select a MindVision Camera for Plastic Inspection?

For plastic and injection molding inspection, MindVision can configure industrial cameras according to product size, defect size, production speed, and system architecture.

Solution 1: Small Injection-Molded Parts

Suitable for:

Small plastic components

Connector housings

Plastic gears

Precision components

Key considerations:

High resolution + macro lens + optimized lighting

Solution 2: High-Speed Production-Line Inspection

Suitable for:

High-speed conveyors

Continuous production

Automated sorting

Key considerations:

Global shutter + short exposure + high-frequency triggering

Solution 3: Multi-Station Injection Molding Inspection

Suitable for:

Multi-angle inspection

Multi-station inspection

Multi-camera inspection

Key considerations:

GigE Vision + multi-camera architecture + trigger synchronization

Solution 4: High-Precision Dimensional Inspection

Suitable for:

Hole measurement

Outer diameter measurement

Profile measurement

Precision plastic-part measurement

Key considerations:

High-resolution camera + telecentric lens + stable lighting

Why Consider MindVision Industrial Cameras for Plastic Inspection?

For machine vision equipment manufacturers, the requirement is not simply a camera. Long-term stability, software compatibility, interface options, and project support are also important.

MindVision's product portfolio includes:

GigE industrial cameras

USB3.0 industrial cameras

High-resolution industrial cameras

Line scan cameras

10GigE industrial cameras

Board-level cameras

Smart cameras

Customized cameras

MindVision's product documentation shows support for interfaces including GigE, USB3.0, 10GigE, CXP, and fiber optics, as well as sensors from multiple manufacturers and OEM/ODM customization.

For long-term automated inspection equipment, this provides flexibility to select different camera architectures according to individual inspection stations and tasks rather than forcing the entire system into one configuration.

Industrial Camera Selection Table for Plastic Inspection

Inspection Scenario Camera Consideration Key Parameters
Small plastic parts High-resolution area scan camera Resolution, pixel size
Flash inspection High resolution + lighting Resolution, contrast
Crack inspection High-resolution camera Pixel resolution, lens
High-speed line Global shutter Exposure, FPS
Multi-angle inspection Multi-camera GigE GigE Vision, trigger
Dimensional measurement High resolution + telecentric lens Resolution, distortion
Small precision parts High resolution + macro lens Magnification
Short-distance acquisition USB3 Vision Bandwidth, USB interface
Long-distance deployment GigE Vision Network architecture
Ultra-high-resolution inspection 10GigE Resolution, bandwidth
Frequently Asked Questions

Q1: How many megapixels does an industrial camera need for plastic inspection?

It cannot be determined from the product type alone. Required pixel resolution should be calculated based on the field of view and minimum defect size, followed by lens evaluation and sample testing.

Q2: Does high-speed injection molding inspection require global shutter?

If the product moves significantly during exposure, global shutter should generally be carefully evaluated.

Q3: Is a higher-resolution camera always better?

No. Lens resolution, lighting, exposure, noise, product positioning, and algorithms all affect final inspection performance.

Q4: What lens should be used for plastic dimensional inspection?

For high-precision dimensional inspection with strict perspective requirements, a telecentric lens can be considered. For general appearance inspection, a fixed focal length or macro lens can be selected according to the field of view and working distance.

Q5: Is GigE or USB3 better for plastic inspection?

There is no universal answer.

For multi-camera, long-distance, network-based production lines, GigE Vision can be evaluated. For compact equipment where the camera is close to the computer, USB3 Vision can be considered.

Q6: Can an industrial camera directly detect injection molding defects?

The industrial camera captures the image, while final defect classification normally requires the camera, lens, lighting, triggering, and image-processing algorithms to work together.

Conclusion

Plastic and injection molding inspection is not simply a matter of selecting a high-megapixel industrial camera. It is a complete machine vision imaging problem.

From determining resolution based on minimum defect size to selecting the shutter according to object speed, every part of the system—including the lens, lighting, GigE/USB3 interface, and triggering—can affect final inspection performance.

A practical solution should follow:

Inspection requirements → Defect size → Resolution → Shutter → Lens → Lighting → Interface → Trigger → Algorithm → Sample validation

MindVision offers GigE, USB3.0, 10GigE, line scan, high-resolution, board-level, and customized industrial cameras for different machine vision requirements. Its product documentation also describes OEM/ODM services covering different interfaces, sensor specifications, and additional functions.

If you are developing plastic appearance inspection, injection molding defect inspection, precision plastic component inspection, or automated vision inspection equipment, the appropriate MindVision industrial camera solution can be selected based on component size, minimum defect size, inspection speed, and installation space.

Reference: https://www.mindvision.ltd/

Why More Megapixels Can Make Machine Vision Worse Industrial Cameras

calliope — Tue, 29 Sep 2026 08:05:52 +0000

More megapixels do not always mean better machine vision.

When selecting an industrial camera, resolution is often the first specification engineers look at. A 20MP camera appears to be a natural upgrade from a 5MP camera, and a 65MP camera seems even better.

But in real machine vision applications, simply increasing megapixels can sometimes make the overall system more difficult to design.

Higher resolution can mean more image data, higher bandwidth requirements, greater processing loads, lower frame rates, more demanding lens requirements, and higher system costs.

The better question is therefore not: “What is the highest-resolution industrial camera available?” but: “How much image information does this inspection actually require?”

This article explains why more megapixels can sometimes work against a machine vision system—and how to choose an industrial camera based on the actual inspection task.

More Pixels Means More Data—Not Automatically More Useful Information A high-resolution sensor captures more pixels. That sounds simple, but every additional pixel becomes part of the image data that needs to be transferred, processed, stored, and analyzed.

For example, if an inspection system moves from a 5MP camera to a much higher-resolution camera, the system may need to handle substantially more data for every frame.

This creates several possible bottlenecks:

Higher resolution → Larger image data → Higher transmission bandwidth → Higher CPU/GPU or processing requirements → Potentially lower system efficiency

This is particularly important for high-speed machine vision. If the production line is moving quickly, the system may benefit more from a camera that provides sufficient resolution at a high frame rate than from a camera with extreme resolution but insufficient speed.

High Resolution Can Become a Lens Problem An industrial camera does not work alone. The final image depends on the complete optical system:

Sensor + Lens + Lighting + Exposure + Camera

A high-resolution sensor can only provide useful additional detail if the lens can resolve that detail. This means that upgrading from a lower-resolution camera to a very high-resolution camera may require upgrading the lens as well. Otherwise, the additional pixels may not translate into additional useful information.

This leads to an important principle: The resolution of the camera should match the optical resolution of the lens and the actual inspection requirement.

For precision inspection, simply buying a higher-megapixel camera without evaluating the lens is not an effective upgrade strategy.

Pixel Size Matters More Than Many Buyers Expect Two cameras can have similar megapixel counts but different pixel sizes. Pixel size affects how the sensor receives light and can influence imaging performance in different application environments.

For example, MindVision's GigE area scan camera portfolio includes products with different pixel sizes and sensor formats. The product selection table covers cameras ranging from compact lower-resolution models to higher-resolution models, with pixel sizes including approximately 2.2 μm, 2.9 μm, 3.0 μm, 3.45 μm, 4.0 μm, 4.5 μm and larger options.

This illustrates why resolution should not be evaluated independently. A practical camera selection process should consider:

Resolution

Pixel size

Sensor size

Field of view

Lens

Working distance

Lighting

Frame rate

together.

When High Resolution Becomes a Frame Rate Problem For a moving target, frame rate can be more important than maximum resolution. Consider a production line where products are continuously moving. If the camera cannot capture enough frames during the movement, increasing resolution will not necessarily improve inspection.

MindVision's GigE camera portfolio demonstrates this trade-off across different products. For example, its GigE area scan range includes:

MV-GE31GC/M — 0.3MP, up to 108.25 fps

MV-GE33GC/M — 0.3MP, up to 160 fps

MV-GE34GC/M — 0.3MP, up to 388 fps

MV-GE40GC/M — 0.4MP, up to 299 fps

MV-GE231GC/M — 2.3MP, up to 40.25 fps

MV-GE300GC/M — 3MP, up to 38 fps

MV-GE500C/M — 5MP, up to 15.25 fps

MV-GE501GC/M — 5MP, up to 24 fps

The point is not that a lower-resolution camera is always better. The point is: Different inspection speeds require different combinations of resolution and frame rate.

Global Shutter Can Matter More Than Megapixels When objects move, shutter type becomes an important consideration. A high-resolution rolling shutter camera may provide plenty of pixels, but moving objects can still suffer from image distortion depending on the application.

For high-speed inspection, a global shutter camera can be a more appropriate choice. MindVision's GigE area scan camera portfolio includes many global shutter models, including models based on sensors such as IMX287, IMX249, IMX265 and other sensors.

Therefore, if the inspection target is moving, the selection process should not stop at: “How many megapixels?” It should also ask: “What shutter type do I need?”

More Pixels Also Means More Bandwidth This is where camera interface becomes important. A high-resolution camera generates a large amount of image data. If the interface cannot handle the required data rate, the system architecture itself becomes the bottleneck.

MindVision provides multiple industrial camera interfaces, including:

GigE

USB3.0

2.5GigE

10GigE

Fiber

CoaXPress

PCIe

The 2026 product portfolio specifically includes GigE area scan and line scan cameras, 2.5GigE cameras, 10GigE area scan and line scan cameras, as well as CoaXPress cameras. For GigE area scan cameras, MindVision supports up to 100-meter stable transmission and multi-camera networking, while hardware image processing acceleration can help reduce host CPU usage.

For even larger data volumes, higher-bandwidth interfaces become useful. MindVision's CoaXPress camera platform, for example, provides an effective bandwidth of up to 4850 MByte/s, while supporting CoaXPress 2.1 and flexible 1-, 2-, and 4-channel configurations.

High Resolution Is Valuable When the Application Actually Needs It The goal of this article is not to argue against high-resolution cameras. High resolution is extremely valuable for applications that genuinely require large amounts of spatial information. Typical examples include:

Large-format display inspection

Semiconductor inspection

PCB inspection

Precision surface inspection

Scientific imaging

Large-area measurement

High-detail industrial inspection

MindVision's product portfolio includes cameras specifically designed for these demanding applications. One example is the MV-PX41000C/M, a 410-megapixel PCIe interface industrial camera. It provides a resolution of up to 24592 × 16704, supports 8 fps full-frame acquisition, uses CoaXPress 2.1, and incorporates TEC plus water cooling to maintain camera temperature stability and reduce image noise.

This is an excellent example of when extreme resolution makes sense: the application requires it, and the complete system is designed to support it.

Sometimes the Right Answer Is a Line Scan Camera One of the most overlooked points in camera selection is that some continuous-material inspection applications are better suited to line scan cameras. For applications involving:

Film

Paper

Metal

Glass

Web materials

Continuous production lines

a line scan camera can continuously capture the material line by line.

Instead of asking: “How many megapixels does the area scan camera have?” the more appropriate questions may be:

What line rate is required?

What pixel size is appropriate?

What material speed must be supported?

What interface bandwidth is required?

Is TDI necessary?

MindVision offers GigE, 10GigE and high-speed line scan solutions. Its 10GigE line scan portfolio includes general models as well as a 256 TDI series, with 5 μm pixels, 10GigE interface and sensor line rates up to 250 kHz/200 kHz depending on configuration. The 256-stage TDI design can reduce the demand for illumination in suitable applications.

This is a good example of why camera selection should begin with the inspection method, not the megapixel number.

The Best Camera Is the One That Matches the Entire System A practical machine vision system can be viewed as:

Inspection Requirement → Field of View → Required Spatial Resolution → Sensor → Lens → Lighting → Frame Rate → Shutter → Interface → Processing Platform → Final Camera Selection

This process is more reliable than starting with a camera catalog and simply selecting the highest megapixel model.

How to Choose the Right Resolution A simple decision framework can help.

Application requirement Priority Camera direction Small defect detection Spatial resolution High-resolution area scan High-speed inspection Frame rate Global shutter high-speed camera Large-area precision inspection Resolution + FOV High-resolution camera Continuous material inspection Line rate Line scan camera Multi-camera factory system Bandwidth + stability GigE / 10GigE Extremely high data volume Transmission bandwidth CoaXPress / Fiber / PCIe Specialized spectral inspection Spectral response SWIR / UV Limited installation space Camera structure Board / Mini / 90° camera 11. MindVision: From Standard Cameras to High-Performance Vision Systems MindVision's product portfolio is designed around different machine vision requirements rather than a single resolution strategy. Its current product portfolio includes:

GigE area scan cameras

GigE line scan cameras

2.5GigE cameras

10GigE area scan cameras

10GigE line scan cameras

USB3.0 cameras

CoaXPress cameras

Fiber optic cameras

PCIe interface cameras

SWIR cameras

UV cameras

Thermal imaging cameras

Smart cameras

3D cameras

Board cameras and modules

Special-structure cameras

OEM/ODM customization

The 2026 product brochure describes MindVision's positioning around high integration, performance, cost-effectiveness, rich interfaces, flexible customization and service.

More importantly, MindVision can customize cameras across different sensor brands and specifications, including Sony, ON Semiconductor, OmniVision, Gpixel and SmartSens, as well as different interfaces and spectral bands. This gives machine vision integrators more flexibility when a standard camera does not completely match the application.

Conclusion: Don't Buy More Pixels Than Your Application Needs The purpose of an industrial camera is not to produce the largest image file. Its purpose is to provide the right visual information for the inspection task.

A 5MP camera can be better than a 20MP camera when the application requires higher frame rate, simpler data transmission, lower processing requirements, or a more cost-effective system. Conversely, a 20MP, 65MP, or even 410MP camera can be the right choice when the application genuinely requires extremely high spatial resolution.

The key is not: “How many megapixels does the camera have?” The better question is: “What does my machine vision system actually need to see?”

MindVision provides industrial camera solutions covering different resolutions, sensor technologies, interfaces, imaging methods and application requirements—from GigE and USB3.0 cameras to 10GigE, CoaXPress, high-resolution PCIe cameras, line scan, SWIR, UV and other specialized imaging solutions. By selecting the camera based on the inspection task rather than simply chasing megapixels, manufacturers can build machine vision systems that are more balanced in image quality, speed, reliability, integration and total cost.

Reference: https://www.mindvision.ltd/

Detection Accuracy Failing? 9 Troubleshooting Directions from Camera to Algorithm (Part 1: Hardware & Physical Layer)

calliope — Tue, 22 Sep 2026 07:47:15 +0000

If you work in industrial quality inspection, you've probably experienced this frustrating moment: the system runs perfectly in the lab, but once deployed on the production line, it "crashes" — or the equipment passes acceptance testing flawlessly, yet three months later the miss rate starts to spike.

In machine vision, detection accuracy is never determined by a single link — from camera selection to algorithm parameters, any weak point can cause the entire precision to "collapse."

This article is divided into two parts, presenting 9 troubleshooting directions based on real project experience. Part 1 focuses on hardware selection and the physical environment (Directions 1–6), helping you quickly locate the physical-layer root causes of accuracy issues.

Direction 1: Camera Resolution — Is Pixel Accuracy Sufficient?

This is the most fundamental yet most overlooked issue.

Pixel Accuracy = Field of View ÷ Camera Resolution. The industry-standard safe rule is: Pixel Accuracy ≤ 1/8 of the Tolerance. That is, if your inspection tolerance is 0.1mm, the pixel accuracy must reach at least 0.0125mm/pixel.

For example: a connector housing inspection project required a tolerance of ±0.03mm. The team used a 65MP camera (9344×7000) with a 42.95×32.18mm FOV, achieving a theoretical pixel accuracy of 0.004597mm/pixel — well above the requirement — and the final measurement repeatability reached 0.01mm. Conversely, if resolution is insufficient, tiny defects are simply "invisible" in the image — no matter how powerful the algorithm is, it's useless.

⚠️ Trade-off Note: Increasing resolution comes with reduced frame rates and longer processing times. If the line speed is high, simply boosting resolution may cause the processing to fall behind the takt time — a trade-off between "inspection accuracy" and "line speed" is necessary.

Troubleshooting Tip: First calculate whether your current pixel accuracy meets the 1/8 tolerance requirement. If not — upgrade to a higher-resolution camera, or reduce the FOV.

Direction 2: Shutter Type — Global Shutter or Rolling Shutter?

Correction Note: This section replaces the original "CCD vs CMOS" discussion. Currently, CMOS occupies over 90% of the industrial camera market, and CCD has largely exited the industrial vision space. For industrial inspection scenarios, agonizing over "CCD or CMOS" is pointless — just choose CMOS. The real critical decision point is the shutter type.

Industrial camera shutters fall into two categories: Global Shutter and Rolling Shutter.

Rolling Shutter works by exposing line by line — the sensor scans from top to bottom, with tiny time differences between rows. This works fine for static objects, but when capturing high-speed moving objects, it produces the "Jelly Effect": moving workpieces appear tilted, stretched, or distorted, directly causing measurement errors.

Global Shutter, on the other hand, exposes all pixels simultaneously, instantly "freezing" moving objects and capturing images that match the actual physical shape — it's the standard for industrial dynamic inspection scenarios.

Major industrial camera manufacturers (such as Hikrobot, Daheng Image, Basler, etc.) all have mature global shutter CMOS product lines, with Sony's Pregius series and Onsemi's PYTHON series being the most typical representatives.

How to tell if you chose wrong? If your images show tilted edges on moving objects, wheels turning into ellipses, or characters becoming stretched — that's the classic "Jelly Effect," and you need to switch to a global shutter camera.

Troubleshooting Tip: Capture a calibration target moving at constant speed and observe whether straight lines become curved or slanted. If so, decisively switch to a global shutter camera.

Direction 3: Lens Distortion & Calibration — Poor Calibration, All Effort Wasted

Real lenses always have distortion — primarily radial distortion and tangential distortion. The geometric distortion error of ordinary industrial lenses directly alters the true size and shape of the target image in high-precision measurement scenarios.

There's a saying: "Poor calibration, all effort wasted." Calibration is not just a routine procedure — the accuracy of the calibration target (ceramic targets have better thermal stability than glass ones), ambient lighting during calibration, and the number and angles of calibration images all affect the final result. After obtaining distortion coefficients through the calibration algorithm, the software must perform distortion correction. If the calibration step is sloppy, all subsequent measurements will carry systematic bias.

Troubleshooting Tip: Re-run camera calibration and check whether the reprojection error is within a reasonable range (typically <0.1 pixel). For high-precision measurement scenarios, prioritize telecentric lenses — they have minimal distortion and are the top choice for precision dimensional measurement.

Direction 4: Lighting Selection & Stability — The "Foundation" of Image Quality

Lighting selection and layout is the part of vision system design that requires the most professional experience. Coaxial lighting is suitable for detecting surface flatness anomalies, oblique lighting highlights surface concave-convex defects, and backlighting is suitable for detecting holes and transmission features. Insufficient contrast caused by incorrect lighting selection is the single most common cause of high miss rates in vision inspection.

For example: a project used red ring light to detect fine scratches on a metal surface, but the scratch direction happened to be parallel to the ring light's illumination direction, resulting in zero contrast. After switching to bar light, the scratch features immediately became clearly visible, and the detection rate jumped from 60% to 98%.

A more subtle issue is light source degradation. Industrial LEDs experience luminous decay over time — brightness gradually decreases, typically dropping 10% to 20% within thousands of hours. If the system lacks closed-loop brightness control or regular verification mechanisms, image brightness slowly "dims," and the contrast between defects and background also degrades.

Troubleshooting Tip: Use a standard gray card to check whether current brightness matches the debug setting. If brightness has dropped, either replace the light source or implement software brightness compensation (such as auto-exposure or gain adjustment).

Direction 5: Lens & Protective Glass Contamination — The Invisible "Filter"

Oil mist, dust, and moisture in factory environments accumulate on lenses or protective glass, forming an uneven transmission layer. This contamination isn't always visually obvious, but in high-sensitivity inspection tasks, it's enough to blur defect images or introduce artifacts.

A detail many overlook: upward-facing lenses accumulate more dust than downward-facing ones. When designing camera mounting, if you can install cameras facing downward or at an angle, it significantly reduces dust settlement. Meanwhile, whether protective glass can be quickly replaced, and whether cameras need protective housings — these design details are rarely mentioned during project delivery, but over a 2-3 year operational cycle, they become critical factors affecting system reliability. For environments with heavy oil mist, adding compressed air purging or oil mist filters is a more proactive protection solution.

Troubleshooting Tip: Wipe the lens and protective glass with a clean lint-free cloth and check if accuracy recovers. If yes, contamination is the root cause — consider adding a protective housing or adjusting the mounting orientation.

Direction 6: On-site Ambient Light & Vibration — The "Gap" from Lab to Production Line

This is the most classic "lab champion, production line bronze" problem.

Here's a real case: a vision inspection system based on Jetson Orin NX achieved 99.8% detection accuracy in the lab, but after mass production deployment, the false positive rate suddenly jumped to 15%-20%. On-site tracking revealed three causes:

① Ambient Light Interference: Sunlight streaming through production line windows caused reflective areas on metal parts to be misjudged as defects, with an abnormal false positive peak between 9-11 AM daily. The lab used an integrating sphere uniform light source (1000±50 lux), while instantaneous illuminance near the production line windows reached 12,000 lux.

② Vibration Blur: Conveyor mechanical vibration induced motion blur — motor start/stop shock peak acceleration of 0.8g, creating a smearing effect in the motion direction, with maximum blur kernel reaching 7 pixels.

③ Material Batch Color Variation: The new supplier's plastic parts had a color temperature deviation ΔE > 8 (lab samples had ΔE < 3).

Solutions included: adding linear polarizers to reduce specular reflection by 60%, using narrow-band filters to suppress ambient light interference, enabling HDR mode to expand dynamic range, and adding magnetic damping mounts to attenuate high-frequency vibration.

Troubleshooting Tip: Add extreme condition tests during acceptance testing — test when adjacent equipment runs at full power, and test separately during day and night shifts. If these aren't tested during the debugging phase, they become "time bombs" after mass production.

Preview of Part 2

In Part 1, we started with hardware selection and the physical environment, covering 6 troubleshooting directions: camera resolution, shutter type, lens calibration, lighting selection, lens contamination, and on-site environmental interference.

Part 2 will move to the software and algorithm layer, focusing on image preprocessing, threshold tuning, and training sample coverage (Directions 7–9), along with a complete quick troubleshooting checklist.

Stay tuned.

Reference: https://www.mindvision.ltd/