Camera

• Resolution: 720 х 576 (CIF4), 360 х 288 (CIF1) is also allowed to use. Increasing the resolution above CIF4 does not improve the operating quality of the recognition algorithm.
• Frames per second: 6 or more.
• Color: color or greyscale.
• No camera jitter is allowed.

Illumination

• Best recognition results are achieved under moderate illumination. If the scene is under- or over-illuminated, the recognition accuracy may drop down.
• Sharp changes in illumination may lead to improper operation of analytics.

Scene and camera angle

• Vertically downward position of the camera is the best for the purpose. The closer to vertical, the more accurate the estimation.
• Camera FOV dimensions: minimum 3x3 m (6x6 humans), optimal 4x4 m (8x8 humans), maximum 8x8 m (16x16 humans).
• The background should be primarily static and should not undergo sudden changes.
• Reflective surfaces and harsh shadows from moving objects can affect the quality of analytics.
• Analytics may not work correctly if there are periodic movements of the background objects in the camera FOV (leafage, TV screens, etc.).

Images of objects

• Image quality: the image should be clear, with no visible compression artifacts.
• Dimensions of a human in scene: bounding rectangle has to occupy from 0,25% to 10% of the frame area.