The requirements for cameras when you work with the People counter detection module are given in the following table.
Camera | - Resolution: 720x576 pixels (CIF4); the use of 360x288 pixels (CIF1) is acceptable. If you zoom in over CIF4, it doesn't improve the quality of algorithm recognition.
- Fps: at least 25.
- Color: only color cameras.
- Camera shaking is absent
|
Lighting | - The best detector operation is achieved at medium lighting. In conditions of insufficient (night) or excessive (light-striking) lighting, the quality of algorithm operation can be reduced.
- Sharp lighting changes can lead to the short-time improper analytics operation
|
Scene and camera angle | - The best position—a camera "looks" at the scene vertically down. The better you perform this requirement, the more accurate the received estimation.
- Sizes of camera field of view: 2x2 m is minimal, 4x4 m is optimal.
- Background is static and isn't changed sharply.
- There are no moving objects except people in the recognized area.
- Analytics can work improperly on reflecting surfaces and in the case of sharp shadows from moving objects.
- Analytics can work improperly if there are periodic movements of background objects (trees, working TV, and so on) in the camera field of view.
- People overlapping by static objects is minimal (columns, trees, and so on)
|
Object images | - Image quality: the image must be clear, without visible defects from the compression algorithm.
- Permissible person size (the area of a rectangular track around a person) as a percentage of the frame area: from 10% to 60%
|
Other | - People must not move by continuous flow, but groups of several people are counted properly
|
