Introduction
Industrial design for a camera usually starts with the form: sketches, renders, the first printed housing. Then it turns out the module doesn't fit, autofocus misses, and the housing runs hot. Every one of those findings costs another iteration.
At PieSoft we build custom cameras, and on a recent project we went in the opposite order. The task was a compact wearable camera. Before any sketches, we put together a test bench: a development board with a camera-specific SoC and an off-the-shelf camera module with autofocus. While the engineers were writing the driver and the autofocus, the bench answered the questions that directly determine the housing.
Below are five of those decisions. They are not tied to our device and apply to any compact camera.
1. The Processor Sets the Lower Bound on Size
The more separate chips on the board, the larger the board and the housing. A camera-specific SoC gathers an ISP, a hardware H.264/H.265 encoder, a MIPI CSI-2 input, an NPU and RAM into a single package. External memory and a separate image processor are not needed.
For design this means that choosing the SoC is already a decision about the size of the device. It is worth making together with the designer rather than handing over finished. Once the processor is chosen, the form is then determined by the battery and the camera module.
2. The Camera Module Sets the Thickness, the Ribbon Cable Sets the Layout
If a device moves along with a person, the distance to the subject changes constantly, and autofocus becomes unavoidable. We took a ready-made module where the sensor, the lens and the focus actuator are assembled together. A separate sensor with an M12 or C-mount lens would have come out larger.
The thickest point of the housing almost always falls in the lens area, and that thickness is set by the module. The ribbon cable, though, can be narrow. Our sensor runs in 2 lane mode: out of the connector's five differential pairs only three are needed, plus I2C and power. The lens does not have to sit directly above the board, and the designer gains freedom to distribute the volume.
3. What Ends Up in Front of the Lens Matters
Our primary autofocus loop is phase detection (PDAF): the sensor itself reports how far and in which direction the frame is defocused. It needs texture. We pointed the camera at a flat wall, and all 192 zones of the PDAF grid returned zero confidence. The lighting was fine at the time. The fallback contrast autofocus then engages, but it is slower.
This is a housing requirement that came out of the firmware. If a smooth surface regularly ends up in the frame or next to the lens, whether a wall, clothing or an element of the housing itself, phase detection autofocus will fall back to the slower mode more often. The lens mounting angle and the way the device is attached or rests are better checked on the bench before the final form.
4. Software Handles Part of the Motion
A wearable camera moves along with a person. We capped auto exposure at 1/100 s so the frame would not blur. The autofocus service reports not only lens position but also state: moving, stopped, focused. In timelapse mode the camera captures not on a timer, but when the lens has settled and the image is sharp.
For design this means the form does not have to absorb motion entirely. The firmware takes on part of the work. A designer can sacrifice less of the appearance for the sake of stabilization, knowing in advance that the software will wait for a sharp frame on its own.
5. Resolution Runs Into the ISP, Not the Sensor
The full frame is nearly 12 MP. The ISP accepts no more than 5 MP at its input. We work in a mode with 2×2 binning, and that is the maximum that can realistically be fed to this ISP.
Before choosing a larger module for the sake of megapixels, it is worth checking that the platform will accept them. In our case it will not, and that is good news for the housing: the extra resolution does not have to be paid for in volume.
Conclusion
None of these decisions is about aesthetics, yet each one constrains the form. If they are checked on the bench before the sketches, the designer works with known dimensions and requirements rather than learning about them on the first housing prototype. That is what we did on the project: software and hardware ran in parallel, and by the time work on the housing began, the module, the autofocus and the sensor mode had already been verified.
At PieSoft we design cameras end to end: sensor, driver, board, housing and software. If you are building a compact device with a camera, write to us: piesoft.com.
Top comments (0)