Computer Vision Analysis Of Leg Movements In The Man-machine Interface

Vision based body movement analysis is a new field in computer vision. Its basic idea is to use a computer to anlysis and recognize human body movement and extract useful information. The fundenments of this technology are the object detection, tracking and recognition. In this dissertation, the research subject is a special case of human body movement analysis, namely, lower limbs analysis and its application in human computer interface.We first reviewed the background subtraction based moving object detection algorithms and the MeanShift tracking methods. Then a set of new algorithms are proposed to meet the special requirements of lower limbs analysis, which include the detection of two legs, the extension of MeanShift tracking to the foreground projection tracking. Finally, a demo system shows the possible applications of our proposed algorithm in human comptuer interaction.The main contribution of this dissertaion can be summarized as follow:Closely analized the color space and similarity metric selection problem, proposed a new criterion to compare the performance of the 8 color space methods. Summarized and analized the MeanShift tracking algorithm, disscused and tested its improvement on bandwidth adaption.Proposed a detection algorithm that can locate the two legs seperately.Extended the MeanShift algorithm to the foreground projection tracking, proposed a replusive factor that enabled the tracker to simultaneously track two legs. By incorporating the Kalman filter, the tracker can then adapt to fast movement.Implemented the proposed algorithms on a software platform. Built a demo system to show the application of our methods in Human Computer Interaction.