Moving Human Tracking In Video And Performance Evaluation

Video tracking system, one essential subsystem of modern video monitoring system, is widely used in video monitoring and security defense, due to its benefits in concealment, direct-spotting and anti-jamming. Video monitoring systems are based on target detection and video tracking, both of which are needed in related advanced post-processing such as target classification and behavior understanding. Moreover, human body's flexibility and human motion's complicacy make the detection and tracking of human body problem, especially the multi-body tracking problem more intractable.It is the purpose of this paper to investigate the probability hypothesis density(PHD) particle filter algorithm, aiming to solve the single-body tracking and multi-body tracking problems in video images. First, the state-of-art moving target detection and tracking methods are reviewed and the Lehigh Omni-directional Tracking System (LOTS) is introduced and validated by simulation experiments. Second, the four classical moving target tracking methods are presented together with simulations, including template match algorithm, mean-shift algorithm, histogram algorithm and particle filter method. Third, the PHD particle filter method based on PHD ,which is using angles as the state vectors,is proposed. Finally, evaluation results algorithm performance analysis in detail are obtained by conducting simulation experiments on the ViPER evaluation platform. It shows that the proposed method works very well in single-body tracking in complex environment. In addition, the proposed method can be applied in solving multi-body tracking problems. Even in case the targets meet each other sometimes, it still works.