Camera

• Resolution: 720х576 (CIF4), 360х288 (CIF1) is also allowed. Increasing the resolution above CIF4 doesn't improve the performance of the recognition algorithm
• FPS is 25
• Color: only color camera
• No camera shake

• The best performance of the detector is achieved under moderate lighting. In conditions of insufficient (night) or excessive (light-striking) lighting, the algorithm performance can decrease
• Abrupt changes in lighting can lead to short-term incorrect operation of analytics

Scene and camera angle

• The best position is the camera looking down at the scene. The better this requirement is met, the more accurate the estimate is
• Dimensions of the camera FOV: minimum 2x2 m, optimal 4x4 m
• The background is mostly static and doesn’t change abruptly
• There are no moving objects in the counting area except for people
• Analytics can work incorrectly on reflective surfaces, and if there are sharp shadows from moving objects
• Analytics can work incorrectly if there are periodic movements of background objects in the camera FOV (trees, TV is on, and so on)
• People are minimally obscured by static objects in the scene (columns, trees, and so on)
• People must not move in solid stream. Groups of a few people are counted correctly