Research On The Algorithm About Target Tracking Based On Image Processing

Target Tracking is a major research direction of computer vision research. It is a highly integrated discipline which relates to computer science and engineering, signal processing, statistics, cognitive science and artificial intelligence and many other fields of advanced technology. The driving force of the research on target detection and tracking comes from the needs of the society, just as the development of all the other technologies. This thesis chooses the PF as the main research object, proposes a new approach with two-step dynamic adaptive resample process, and implements a tracking system based on the new approach.The major work of this thesis is as follows.This thesis compares and analyzes the standard particle filter algorithm (SIS) and the variant improved PF, e.g. regularization (RPF) particle filter, auxiliary sampling particle filter (ASIR). Then this thesis studies the TLD algorithm and on this basis implements a more efficient tracker. After discussing the shortage of the PF and the existing improvement method, this thesis concludes the theoretical basis of the new method.Detection step introduces in the resample process. First, the new system creates the tracker and use the tracker to determine the target location and the degree of confidence of the detection. The output of the detection was used to guide the re-sampling process, e.g. determine the dynamic resample range of particle sampling R and the number of samples N. The method improves the efficiency of particle sampling process, and gives the specific forms of implementation of the improved algorithm. Finally, the thesis designs experiments to verify the new method. The results prove that the algorithm has a higher real-time than the each single method, improves the accuracy of target tracking, and effectively reduces the cost of the algorithm.