Head detection is a core AI feature and an initial phase executed in a Vision system.
Head detection execution is independent of the main subject's head orientation, the very reason why the system is able to complete the detection with various head appearances.
Basically, in order to detect the head, the subject has to be positioned in the scene (visual field of the camera).
Once a subject appears in front of the camera, the system makes a thorough analysis of a full frame in order to detect the head of a subject.
Each head detection, as well as face detection, is done on each frame that the camera captures.
Conclusively, the higher the framerate the larger the number of detections. The detection success rate gets lower the further away the subject is from the camera. Head detection is not connected to face recognition, except for providing a foundation for face detection. People count and heat maps, additional features of this system are done based on head detection.
The Vision System uses a neural network consisting of a feature called Single Shot Multibox Detector, the method shown to be quite easy to implement and more precise compared to others