Heuristics and Prior Knowledge

It is almost always a good thing to use as much prior knowledge about the task ahead as possible. For face detection we know that the head is located on top of the body, and that a human normally walks upright and so on. We have formulated the following heuristics to capture some of this prior knowledge we have of the task:

• If there is a large moving object in the image, there may be a human present

• If the movment in the upper part of the moving object is larger than a threshold, this may be a the top of the human, thus a face

• If a face is detected in a subframe of size $40\times40$pixels or larger, the human detected is working on this PC, not just passing by or standing in the background

Ths system also has to deal with situations where a human may not be present, or present and not moving. In these cases we use a global boolean variable to decide if a human is present or not. For instance, if a human is detected and does not move for a long time, the system asks for input within a few seconds. If no input is given, the system has probably failed to detect the human leaving, and changes the global variable to indicate no human present.