Research On Human Body Motion Video Tracking Based On Adaptive Particle Filter

All kinds of cheap devices such as digital cameras and personal computer are used to capture a human's motion without any sensors or markers attached to the body, which has becomes a hot topic of computer vision recently. Video-based Human body Motion capture often becomes difficult due to the complexity and disorderliness of the target action. Although video-based human body motion capture has been widely researched and many effective algorithms have been proposed, there are still a lot of difficulties in capture the human body motion based on complex movements.In this thesis, the research is focused on the video-based capture problems of complicated movement. Since these problems are always nonlinear and non-Gaussian, particle filter algorithm based on the probabilistic model is used as a capture framework. Although the particle filter algorithm can fulfill some common capture task, there still remain some difficulties in video-based tracking of the human body motion. Diving videos capture with the original particle filter may have two major problems. First, the observation model can't adapt to the change of athlete body appearance. Second, the fixed noise variance in the kinematics model usually fails in predicting the rapid change of athlete location. Here, an adaptive particle filter algorithm to tackle these problems is presented, which introduces a kind of adaptive observation model under the particle filter frame. In addition, it can choose noise variance and the particle quantity according to capture errors and the changing scope of motions adaptively.In this thesis, lots of human motion videos are tested by capture and a comparison with the traditional particle filter algorithm is done. Experimental results show that this algorithm can not only fulfill general human foot video tracking, but also complete the video-based capture of complicated movement. Furthermore, this algorithm has a relatively lower tracking error rate.