Research On Object Detection And Tracking Based On Nonparametric Kernel Density Estimation Background Modeling And Mean Shift

Video object detection and object tracking has been widely used in variousfields. Based on the nonparametric kernel density estimate in backgroundmodeling and Mean Shift target tracking, this thesis puts forward new methodsto improve object detection and object tracking in real-time performance,robustness and accuracy.The main contributions of this thesis are summarized as follows:1．Due to serious time-consuming in background modeling calculating allthe pixels in the map, in this thesis an improved algorithm is proposed usingregional pixel’s information and smoothing method to reduce calculation. As aresult, our algorithm can obtains favorable tracking experimental in real-time.In subsequent noise reduction section, with the help of neighborhood pixels’connectivity noises are effectively suppressed, and the details of thesegmentation result are well kept.2. Traditional Mean Shift object tracking method has poor tracking resultswhen background and foreground have similar color distributions. Base onTraditional Mean Shift object tracking, this thesis proposes an improved backprojection using back projection map. The experimental results show that ourtracking algorithm improves the accuracy and robustness.