|
Video multi-target tracking is the field of computer vision is one of the core issues , which combines image processing, automatic control , pattern recognition , artificial intelligence and other disciplines of advanced technology, is widely used in military , video surveillance, traffic control and human machine interaction , and other fields , and has broad application prospects and great economic value. Based on static, single camera video sequences , respectively, multi- target tracking in video video object detection, filtering algorithms and multi-objective association conducted extensive research and simulation analysis . First, the introduction of inter-frame difference method and a simple background subtraction , and focuses on the Gaussian mixture background modeling method using two color space for multiple thresholds for shadow removal , has been very good detection results. Secondly, the introduction of the adapted Gaussian linear Kalman filter algorithm , focusing on learning to adapt to nonlinear non-Gaussian particle filter for the subsequent video track laid a theoretical foundation . Again, the introduction of multi- target tracking data association algorithm for multi-objective characteristics of the video , choose to Euclidean distance and the combination of color histogram distance target association method, the trajectory and determine a reasonable match conflict from happening. Finally, on the basis of previous knowledge presents a particle filter based adaptive filter tracking algorithm associated with the first multi-objective judgments based on conflict situations , there is no conflict characterized by color selected observation model , the selection of the spatial conflict integration of information and color characteristics of observation model , a reasonable allocation of computing load conditions to deal effectively with a multi- block between the objectives , complex motion separation and merger cases .
|