Research Of Multi-Target Tracking Algorithm Based On Main Detect And Multi-Feature Parameters

Multi-target tracking, which belongs to computer vision, has become an independent field crossed with many other subjects. With the development of science and technology, multi-target tracking has been used widely, for instance, the long-standing military radar tracking system and kinds of video surveillance management systems we are using in traffic, security, prison, bank, school. Nowadays, more and more attention attached on multi-target tracking has brought plenty of creative achievements, however, higher demand has also been put forward, especially cases of considering people as the tracking objects being desperate for better algorithm to solve more complicated problems. Therefore multi-target tracking is endowed with a promising future.According to the processing flow, all steps of multi-target tracking are introduced in detail in this pager. First of all, the main methods of target detection and target tracking, the two stages of multi-target tracking, have been put into research in theory. What’s more, considering advantage and disadvantage of those methods, this paper presents two algorithms for each stage of multi-target tracking, and meanwhile the algorithms will be tested and verified through experiments and comparison. The main work in this paper is concluded as follows: (1) Put forward the main detect algorithm. Obviously, when acquiring separated targets, it is quite difficult for the traditional target detection algorithms to acquire accurate target templates. According to the above-mentioned problem, this paper combines the result of foreground detection with statistical approach, takes the main of targets acquired through statistics of foreground pixels as the target templates, and then gets rid of details that don’t need to be considered in the tracking stage as well as unstable parts which are liable to causing interferences. Finally target templates are acquired absolutely and used to get feature parameters in the tracking stage. As it is proved by experiments, main detect algorithm does quite well in acquiring the tracked target template.(2) Put forward the multi-target tracking algorithm based on main detect and multi-feature parameters. There exist three crucial dilemmas in multi-target tracking, which are tracking with targets being not shielded, tracking with targets being shielded by something fixed, and tracking in the case of multiple targets’being shielded mutually. The multi-target tracking algorithm, combining multi-feature parameters with particle filter, is designed for these three crucial dilemmas. In comparison to traditional target tracking algorithms based on SIFT feature, it has been proved by lots of experiments that not just can the tracking algorithm achieve multi-target tracking, but also figure out these three crucial dilemmas perfectly in this paper.