Camshift Based On Multi-feature Fusion For Real-time Visual Tracking

In this paper, by the Kalman filtering and Camshift algorithm, I found the classicalalgorithm can effectively solve the problem of target deformation and occlusion, and achievebetter tracking results in a simple background. However, when the background is complex, orthere are many pixels that are similar to the target in the background, it is easy to fail byclassical algorithms. Because it just consider the color histogram. If we take into account thespatial distribution of the target, the limit characteristics will be added. Through in-depthstudy of camshift algorithm and kalman filtering, I found that the two algorithms in fact havea complementarity. So we combined them and overcome each other’s deficiencies andshortcomings. In many cases, the background and target are mixed together. At this time, it isdifficult to modeling of the target and the candidate target. Then the histogram is divided intoseveral small intervals which are called Bin. The number of the target background ratio aseach bin’s ability to distinguish between target and background. Then we can just find thedifference between the target and the neighborhood. In the pre-video image processing we usefeature extraction method. By feature extraction noise interference can be reduced. We usemulti-feature fusion techniques to deal with visual tracking image，it can get the minimizenoise. In this paper we propose a fusion visual tracking algorithm by combining kalmanprediction with camshift algorithm based on the multi-feature, for automatic Target Detection,By the weighted color histogram and the weighted gray gradient histogram, not only in termsof the robustness of the tracking results were better, but also in the real-time tracking of thetarget it has very good results. It also can meet the needs of real-time distinguished tracking,while reducing the number of iterations in about23%. And it greatly improve the visualtracking efficiency.