Human Segmentation in a Complex Situation Based on Properties of the Human Visual System

Segmenting multiple humans is a challenging problem in complex situation. This paper proposes a human segmentation method by fusing motion, colour and stereo cues. Making use of the psychophysical and neurophysiological conclusions about human recognition, the algorithm mimics the mechanism of HSV to human segmentation. In simple situation, it can segment the human body by using colour and motion cues. In complex situation, especially people overlapping in the image because of the 3D scene projection onto 2D image plane, the difference frames can not segment each object. We add the depth cue to segment the overlapping object. Experimental results prove the algorithm works robustly, adaptively and efficiently