关键词: 深度信息, 均值漂移, 带宽自适应, 颜色直方图

Abstract: The background noise in the candidate object model diminishes the object color characteristic, and induces localization error. To reduce the error, according to the discriminative depth level between the object’s and the background’s, a Mean Shift algorithm based on depth cues assisted and corrected background-weighted histogram is proposed. The proposed algorithm can sufficiently weaken the background noisy interference in the kernel window, enhance the object’s color feature information, and update the kernel size adaptively in due course to reduce the distractive information in the background as the object size becomes small. Experimental result shows the proposed algorithm has fewer iteration number and good localization precision of tracking.

Key words: depth cues, Mean Shift, adaptive kernel bandwidth, color histogram