The Research Of Object Tracking Algorithm Based On Multi-Feature Fusion

With the rapid development of computer technology, object tracking has been widely used in intelligent surveillance, precision guidance, human-computer interaction, intelligent transportation system, etc. It has long been a hot research topic in the field of visual analysis. This topic has a great significance both in the scientific research and practical application.So far, there are still some technical difficulties in the field of object tracking, such as complicated background, all kinds of interference, object scale variation, tracking robustness, etc. In order to enhance the robustness of dealing with all kinds of interference in complicated environment, this paper focuses on how to effectively fuse multi-feature and how to achieve object scale adaptive in the framework of multi-feature fusion. Based on these, a tracking algorithm based on multi-feature fusion is proposed in this paper. The experimental results prove its effectiveness. The main work and results are as follows:(1) With a comparison between the international and domestic research situation of object tracking, four technical problems are analyzed. The pivotal issue is proposed, which is how to deal with complicated background and different interference better. A reference basis is provided for the new method of this paper.(2) According to the advantages and disadvantages of various classical tracking algorithm in different situations, this research focuses on Camshift algorithm and particle filter algorithm. In the framework of Camshift algorithm, object tracking based on color features is simulated. The results show that it can deal with illumination changes and complicated background with little color interference.(3) In order to solve the exposed problems in object tracking based on color features, a new idea about multi-feature fusion in the framework of particle filter is provided. In addition to fusion of color features, edge features and motion features, this paper proposes to select color features or edge features according to their own observation probability. It isn’t only easy to be calculated, but also improve robustness of anti-interference. Experimental results show the proposed method has high accuracy and good robustness.(4) The proposed method can’t deal with scale changes of the object in the video sequences. So it is suggested to integrate scale adaptive method into the object tracking based on multi-feature fusion. According to the theory of Camshift algorithm, moment features are used to adjust the size of the search window. Experimental results show better tracking performance of the improved method.