Weighted Sub-block Mean-Shift Tracking with Improved Level Set Target Extraction

摘要

Mean-shift tracking algorithm is a widely-used tool for efficiently tracking target. However, the background change and shade usually lead to tracking errors and low tracking accuracy. In this paper, we introduce a novel mean-shift tracking algorithm based on weighted sub-block which incorporates the improved level set target extraction. The weight of each sub-block is determined by the similarity of target and candidate sub-blocks, and by the ratio of the target sub-block and overall areas. The target sub-block area is calculated by the means of the narrow band level set combined with a compromise to improve extraction accuracy and operating efficiency. Both of RGB color information in the target region and the pixel's position information are taken into consideration while describing the feature model of target and candidate region inside each sub-block. Experimental results demonstrate the method's success for tracking of targets with background change and shade during the dynamic scene, where the basic mean-shift tracking algorithm fails. The proposed method has better tracking performance with higher tracking accuracy and adaptability.