Moving object segmentation based on new likelihood functions

Accurate segmentation of moving objects from a video sequence is still a difficult task. A moving object segmentation method is proposed in this paper to deal with the segmentation splits and defects. First, it is claimed that the confusion point is one reason for the segmentation inaccuracy, and corresponding solution is also presented. According to the solution, new likelihood functions are proposed to compute membership probabilities, which are then used for final segmentation within an energy minimization framework. Unlike related algorithms which compute membership probabilities using kernel density estimation, the proposed method models the membership probabilities as functions with kernel density estimation as the independent variable. Experiments show that improved results are generated by the proposed likelihood functions.