The megapixel race made night performance worse
The single most useful thing you can know about choosing a security camera is this: putting more megapixels on the same size sensor makes the night image worse, not better.
A 4 MP image on a 1/3-inch sensor gives roughly 2.0 µm per pixel. An 8 MP image on that same sensor gives about 1.4 µm. Smaller pixels collect fewer photons. Less signal means more amplification, and amplification is exactly what noise is. So the camera with the bigger number on the box frequently produces a grainier, less useful night image than the previous generation did.
If your priority is identifying a face or a plate after dark — which it usually is — then sensor size and aperture beat resolution every time. Compare these instead:
Pixel pitch. Anything from 2.0 µm upward is reasonable for low light; below roughly 1.6 µm you will see it immediately.
Aperture. An F/1.0 or F/1.2 lens admits dramatically more light than a common F/2.0. It is one of the few specifications where the difference is genuinely visible rather than marketing.
Number used pixels versus sensor format. 'A 1/1.8-inch sensor at 4 MP' is a far stronger statement than '8 MP' with nothing else.
Where the IR LEDs should be, and why it matters
Most failures blamed on "weak night vision" are actually reflection problems.
When the IR LEDs sit directly around the lens behind a single flat window, some of their light bounces straight off that window — or off dust, a spider web, or rain droplets — back into the lens. The result is a milky haze, sometimes a solid white-out of one corner. The same happens if the camera looks along a wall, down a soffit, or anywhere there is a nearby surface to bounce off.
Two things fix it:
Physical separation between the illuminator and the lens, so the light cannot reflect straight back. Better domestic designs move the LED ring away from the optical axis, or place them under a separate window.
Moving the camera. A camera mounted under a deep eave that sees mostly that eave will always suffer. A few degrees of change often matters more than any product swap.
This is also why through-glass indoor mounting never works. Infrared reflects from glass, not passes through it usefully. The image either whitens or simply shows the room's own reflection. If you need to look through a window, mount outside the glass, or accept that night mode has to be off and rely on ambient light.
IR-cut is mechanical or it is not real
During the day the sensor must see visible light only, otherwise foliage and skin render wrong. At night it must see infrared, or there is no image. A mechanical IR-cut filter physically slides into and out of the optical path to do this. Cameras without it guess in software, and you can tell — for twenty minutes at dusk and again at dawn, everything takes on a flat pink cast.
You can hear the mechanism operate: a small click when it switches. If you never hear it, you have a software-only implementation, and you should expect the colour shift.
Wide dynamic range, and the vendor numbers that lie
Entrances, driveways viewed from a windowed room, and anything facing a street lamp are always backlit. In those scenes you need WDR.
Be sceptical here. "Digital" or "DWDR" generally applies a tone curve that lifts shadows, leaving highlights still blown out. Genuine multi-exposure WDR combines a short and a long exposure per frame. A real figure around 120 dB is meaningful; below about 100 dB you should assume it is mostly marketing. Test it pointed at a doorway with daylight behind before you believe it.
Storage maths, done once properly
Video storage is bitrate multiplied by time, and resolution is only indirectly related.
A rough working figure for H.265: a 4 MP stream at around 2–3 Mbps consumes roughly 21–32 GB per day per camera. Three cameras, thirty days retention, is therefore on the order of 2–3 TB — not the 1 TB drive that came in the kit.
Things that move the number:
H.265 versus H.264. Roughly a 40–50% saving for similar perceived quality. Worth having, provided everything downstream can decode it.
Frame rate. 15 fps instead of 25 fps cuts nearly half the storage and changes very little for general surveillance. Most people never notice.
Motion-triggered versus continuous. The largest single lever, but note it works against AI detection: most person and vehicle detection needs a continuous low-res stream to analyse, so the saving is not as clean as it looks.
Bitrate is dynamic. IR at night raises noise, and noise does not compress — so night footage can use more space than daytime despite looking worse. Size for the worst case.
Use high-endurance cards if you use cards at all. A standard microSD doing continuous writes in a hot enclosure is the least reliable component in most installations, and it fails silently — you discover the gap when you need the footage.
The two questions to ask before purchase
Not the marketing questions. These two:
Does it expose RTSP or ONVIF to a local recorder? If not, you are locked to one vendor's app and one vendor's cloud. This matters more than any image feature.
What happens if the internet goes down? Some cameras stop recording entirely when they cannot reach their own cloud. A camera that keeps writing to local storage during an outage is the difference between having evidence and having a story.
We build CCTV and surveillance platforms for OEM and ODM programmes, so read this as the commissioning list rather than neutral advice. Current platforms are on the CCTV camera manufacturing page, and the video intercom range covers the entrance side of the same projects.
The short version: judge a camera by pixel pitch and aperture rather than the megapixel number, keep IR away from the lens axis, insist on a mechanical IR-cut, and size storage for night bitrate.
XingLiao builds CCTV camera platforms, video intercom systems and smart lock manufacturing for OEM and ODM programmes. See XingLiao for platforms and specifications.
Top comments (0)