The Improvement of Mean-Shift Algorithm in Target Tracking

Abstract

The bandwidth of kernel function is invariance in Mean-Shift tracking process which led to the problem of actual process of tracking failure. To solve this question a method is proposed to combine object contour, RGB color histogram and Mean-Shift tracking algorithm. In each frame the contour and color feature of target object are extracted to as a model. Meanwhile the size of contour is used as the bandwidth for kernel function in the next frame. Based on this Method the accuracy of Mean-Shift tracking is improved and the error of Mean-Shift tracking is reduced.