The main recommendations concerning camera choice and installation for the Face Recognition mechanism:
1. Camera must be connected to broadcasting device through USB;
2. Video resolution should be not smaller than 1280×720px (720p), frame rate should be no less than 25 frames per second;
3. The lack of autofocus;
4. It`s necessary to create optimal conditions for camera in order to work correctly.
Recommendations for correct camera operation:
We recommend to use our solutions in well-lighted places without light variations, lights and backlights (neither against sun, nor against other light sources).
Targeted use of Face Recognition technology isn`t directed to communication with a crowd of passers-by but with 1-3 people for more than 5-10 seconds.
The detection and recognition mechanics are based on faces (not on shadow figures, "heads" etc.). As a rule, a camera should be installed at the face level (1.5-1.8 meters height), near the display (or built-in the display), pointed at faces. It`s necessary to avoid the angle view (from above or the side).
In case of the built-in cameras for kiosks, laptops etc., the camera should be pointed at user coming close to device.
You can use the utility of image preview (usually it`s supplied with drivers or pre-installed on OS) for exact position of camera.
By camera features:
The Face Recognition mechanism works with RGB cameras. We recommend using cameras that have HD, FullHD-quality video with 25-30 FPS. Cameras must be without noise (depends on lightning and installation circumstances) but with qualified matrix. Otherwise, the Face Recognition quality decreases while the imprecision increases.
Camera must be without autofocus (or with autofocus turned off). It is connected to nettop/computer (broadcasting point) and should be the HID device being available through OS. As a rule, these are USB Webcams or built-in modules working on the same principle.
By number of faces:
There are no explicit limits. This mechanism considers all the faces, including their tracks, identificators and properties. This system detects 6-8 faces at the same time in practice. It`s the maximum number of people fitting in front of the camera. Nevertheless, the main feature is the interaction with 1-3 people, while the others are used for statistics, reports and background audience analysis.