Research On Moving Targets Detection And Tracking In Video Sequences

Involving numerous subjects, moving targets detection and tracking is a hot research in the field of computer vision and signal processing. And it has been applied widely in many aspects such as military, industrial, life and so on.Taking the movement of vehicles as examples,explorations and studies on all aspects of the moving target detection and tracking have been carried out in this paper, and some viewpoint and details have been put forward. In this paper, the main algorithm of movement object detection and tracking has been discussed emphatically. The main cotributions of this dissertation are summarized as follows:1. When background subtraction method been using to detect the movement goal, it is very important for the extraction and reconstruction of background image correctly. After the shortcomings being considered, the time-axis statistical method and the small region processing method are proposed. According to the former experiment, these two methods can be more accurately extract the image'background, and don't require more image samples.2. On basis of pattern plate registration, a method integrating Pyramid multi-resolution and Diamond search method was presented. There are some steps about the main idea of this method. Firstly, we can use the average value Pyramid to process the goal template and the partial regions, which lower the resolution. Secondly, at every image layer, using the Diamond search method for tracking regions of positioning roughly. This method may reduces the search-costing times and raises the search efficiency. Besides, the paper describes the algorithm principle and the track step in details, and has carried on the comparison and the explanation to the experimental result.3. Aiming at the traditional Mean Shift tracking algorithm can not track the window size fixed insufficiency, two kind of improvement algorithms been presented. Firstly, Fuzzy Threshold Segmentation have been used for target tracking with Mean Shift to improve deficiency that the bandwidth of kernel function of Mean Shift is not changeable. The weight of gradation transformation is ascertained by the contrast of the target and background RGB color. Secondly, Fuzzy C-means clustering has been used in the YCrCb colorspace for the target segmentation from the background. According to certain rules,the bandwidth of kernel function is corrected with the statistic of target pels. Experiment results show the two algorithm can track objects accurately, effectively and real time. The new algorithm also overcomes the limitations of the traditional one.This research is mainly realizes through the simulation. After the algorithm optimization and the transplant, the real-time track can apply on high performance hardware system such as DSP.