A Multi-target Tracking Approach Based On Multi-feature

In the modern society, with the rapid development of intelligent video surveillance system, people are paying more and more attention to security. Intelligent video surveillance system will firstly apply image processing and target detection to the video sequence flow, and then identify the target and track the target. Video target tracking is an important part of intelligent video surveillance. Its research involves visual analysis, pattern recognition and other fields and it has a huge prospect of application. Tracking includes single target tracking and multi-target tracking, but the latter request higher robustness, real-time performance and stability.In video multi-target tracking process, the light mutation, and targets occlusion each other make the tracking becomes a key problem. Based on aforementioned problem, the thesis mainly research it from three aspects including target detection, feature extraction and tracking algorithms. Besides, this thesis also put forward a set of solutions. To solve the key problem, it uses multi-feature fusion of moving target color feature, location feature,trajectory feature. Through calculating the similarity of target feature information, which can effectively solve the problem of target lost or wrong target in tracking. This major research approaches in this paper are:This paper proposes a moving target detection algorithm that based on Gaussian mixture model. This paper will first apply the Gaussian filter to video image, and then detect and extract the target area by using the Gaussian mixture model. Finally, do corrosion expansion operation to the binary target image, fill the small and medium-sized hollow of target area, gets rid of the noise and we will get a complete target image.For the sudden changes in light will affect color features of the target and then affect target tracking, multi-feature fusion algorithm is put forward. In the algorithm, we matching tracking the target, by fusing target multi-feature of the color features, edge features, and physical features. The experimental results show that the algorithm can effectively solve the problems.To solve the problem of target occlusion, this paper proposed multi-target tracking algorithm which is based on multi-feature fusion. By fusing target multi-feature, matching tracking the target. The thesis studied the condition of target occlusion and put forward a method to judge the occlusion when it occurred and ended. We carried on the experiment many times in the process of tracking due to the sudden change of light and occlusion. The experimental results show that the algorithm can effectively solve the problems above with high efficiency and good robustness, and it can realize the real-time tracking multi-targets in the video.