Mean-Shift Target Tracking Algorithm Based On The Background Information

In the field of computer vision, target tracking is a challenging research topic. Researchers have done a lot of work on this topic and made many achievements. Among them, the mean shift tracking algorithm has been widely used because of its efficiency, simplicity and other advantages. But traditional mean shift tracking algorithm also face the following problems in practice:background information on targeting, background model and update policies, the color characteristics susceptible to the effects of illumination change.In order to solve the problems of background interference in the mean-shift tracking algorithm, an improved algorithm based on color and texture blending characteristics and background weighted update approach is proposed. The original RGB image is converted to the HSV color space, then color feature is extracted in the H, S channel and texture feature is extracted based on an LBP descriptor in the V channel. Base on this, the color-texture histogram of the object region and background is established. During object tracking, the background region is updated using a weighted update approach according to the Bhattacharyya coefficient. In order to improve the efficiency of the updating background model, the thesis proposes an improved algorithm based on the block background update. The background around the region of the target area is divided into four sub-blocks and the color histogram model of the four sub-blocks is calculated, respectively. The similarity of the lower three sub-blocks model is selectively updated by Bhattacharyya coefficient.In the construction of the target candidate model, object spatial information is used to give the corresponding distance of weight to the candidate target of the pixels.The extensive experimental results show that, compared with the algorithm adopting the full background update approach with color or color-texture features, the improved algorithm makes full use of color and texture features and adopts weighted updated background region, and has a higher level of reliability and robustness and better execution efficiency. Experiments show that the strengthen algorithm converges faster and has better resistance to background interference and noise.