Object Tracking Algorithm Research Based On Particle Filter

The object tracking, which compromises computer technology, pattern recognition,artificial intelligence and other technical disciplines, is an important research focus area in the computer vision, and has board research and application prospects. Currently, target tracking has been widely used in many applications, such as military, the video surveillance, traffic control, human-computer interaction and medical diagnosis. In recent years, the particle filter and sparse expression have been widely used in the field of computer vision, and have made good research results. Furthermore, many excellent research results have been achieved in target tracking. Therefore, a robust object tracking algorithm based on particle filter is proposed in this paper which includes two parts as follows:Firstly, the particle filter tracking algorithm based on adaptive multiple cues fusion is introduced. The tracking algorithm is generally used one single characteristic to describe the target, however, it is difficult to accurately describe the tracking target under the circumstances of complex background, lighting change, and thus the tracking result has instability. In the view of the above problem the combining multiple features is presented in this paper which combines three complementary features: color, texture and edge to accurately describing the target. The weight of each feature is adaptive changed in the tracking process, thus the target can be accurately described in different situations, and it the tracking algorithm becomes more robust.The object tracking algorithm with particle filter based on sparse mixed model is introduced in this paper. The target is generally viewed as a whole in particle filter algorithm.If the target appears to be long time occlusion, tracking results will appear to drift phenomenon or complete loss of the target in the particle filter algorithm. So in this paper we propose a mixed model which exploits both holistic templates and local representations.Sparse represents the local templates via the complete dictionary of the target, and the tracking effect is more accurate by the means of analyzing the local target.Comparing with the existing tracking methods, the experimental results indict that the algorithms proposed in this paper are effective and robust.