Too small a face portrait will affect the cognitive effect, while too large a face portrait will affect the cognitive speed. The minimum pixel for non-professional face recognition cameras is usually 60*60 or 100* 100 or above. Within the specified image size, the algorithm can easily improve the accuracy and reproduction rate. The size of the image reflects the distance between the face and the camera in the actual application scene.
2. Resolution of the image
The lower the image resolution, the more difficult the face recognition system is. Image size and comprehensive image resolution directly affect the camera recognition distance. At present, the maximum distance for a 4K camera to see a face is10m, and that for a 7K camera is 20m.
3. Lighting environment
Overexposure or too dark sunlight are all factors that endanger the actual effect of facial recognition. It can supplement light or filter light from the built-in function of the monitoring camera to balance the harm of sunlight, and it can also use the solid model of the calculation method to enhance the image light source.
4. The degree of occlusion
Images with clear facial edges are best. However, in the actual scene, many faces are blocked by obstacles such as umbrellas, myopia glasses and anti-fog masks. These statistics consider optimizing this problem through the algorithm model.
5. Collection angle
The angle of the face relative to the camera is most suitable for the front. But in actual scenes, it is often difficult to take photos. Therefore, the algorithm model needs to train data including left and right faces and upper and lower faces.