CAJ | 학술논문

基于颜色直方图的传统Mean Shift视频目标跟踪算法，容易受到背景信息的干扰而造成定位偏差。针对此问题，提出了一种基于漂移系数的加权颜色直方图算法。根据核窗口内每个像素提供系数的大小，抑制对漂移贡献小的颜色特征，从而增强目标特征的刻画，并通过抑制背景信息和引入新的目标信息来更新模板。实验结果表明，该算法定位跟踪效果优于背景直方图加权算法，并且迭代次数更少。采用模板更新后，跟踪性能更加稳定。
기우안색직방도적전통Mean Shift시빈목표근종산법，용역수도배경신식적간우이조성정위편차。침대차문제，제출료일충기우표이계수적가권안색직방도산법。근거핵창구내매개상소제공계수적대소，억제대표이공헌소적안색특정，종이증강목표특정적각화，병통과억제배경신식화인입신적목표신식래경신모판。실험결과표명，해산법정위근종효과우우배경직방도가권산법，병차질대차수경소。채용모판경신후，근종성능경가은정。
The conventional Mean Shift video target tracking algorithm based on color histogram is very sensitive to the interference of background noise, lead to the target location error. To address this problem, a weighted background algo-rithm based on shift coefficient is proposed. According to the shift coefficient given by each pixel in the kernel window, the proposed approach can reduce color feature amount making small contribution to mean shift in the color space, enhance the characteristic of the target, update the template promptly by restraining background interference and introduce new target cues. Experimental results show that the proposed method performs better than background weighted histogram algorithm with less iterations, and even more stable with the template update mechanism.