Choosing a camera

Once you have picked a source type, the camera's optical and environmental fit decides whether recognition works in practice. Aim for Full HD (1080p) resolution and a field of view matched to the lane — too wide a lens shrinks faces and distorts them, so a person only a short distance away may already be too small to match. The working target is at least 50 pixels between the eyes at the recognition point; pick the lens and mounting distance that achieve it.

Lighting is usually the hardest variable to control and the most common cause of poor results. Favour cameras with strong low-light performance, keep any light source out of the camera's field of view, and prefer consistent frontal lighting over backlight or harsh side light. Wide Dynamic Range can help in mixed-contrast scenes, but it introduces artefacts that lower passive-liveness scores, so leave it off where liveness matters. Run at 15 fps or more at all times, since low light can quietly drop the frame rate.

Treat power, network, and mounting as part of the decision, not an afterthought. Power over Ethernet (PoE) carries data and power on one cable and greatly simplifies installation, so prefer it where you can. For server-side RTSP, budget around 4 Mbps of stable bandwidth per camera and use constant-bitrate, quality-priority compression. Mount at roughly eye level facing oncoming travelers, keeping head angles modest (roll ±15°, pitch and yaw ±30°) for the most reliable matching.

On vendor choice, the guiding principle is simple: buy reliable hardware from reputable brands with real warranty and support, and avoid low-end cameras — they add latency and perform poorly in low light, which is exactly where biometrics already struggle. A versatile option is a camera that supports both RTSP streaming and an on-edge processing chip, so you can switch source types later without replacing hardware. For the full set of optical, lighting, and mounting considerations, see the camera guide in the VPP platform documentation.

Compatible hardware

For server-side RTSP, any camera streaming H.264 or MPEG-4 will work; among proven vendors, AXIS is a reliable choice, and models such as the AXIS M4216-LV support both RTSP and an on-edge-capable chip, letting you choose either source type with one device. For on-edge processing, use a device from the supported-devices list in the VPP embedded documentation (Ambarella-based smart cameras, or AI boxes using Hailo-8 or NVidia Jetson). Since pricing and availability vary by market, research current options and share any shortlist with Innovatrics for a compatibility check.