针对采用核加权直方图的方法计算目标模板与候选区域目标特征无法实现对运动目标的准确定位这一问题, 提出了一种利用改进背景加权增强直方图显著性的鲁棒Mean-Shift跟踪算法。在传统Mean-Shift的框架下, 通过计算目标和背景特征直方图bin值, 得到两者特征显著性大小, 将其代入传统相似性度量中, 定义新的背景加权系数, 进而更好地提高目标与背景的区分度, 减少背景信息对目标定位的干扰。通过算法改进前后的实验结果以及跟踪误差和正确跟踪率的比较发现,跟踪效果得到明显改善。

Considering that it is difficult to locate the moving objects accurately from the clutter background by using the weighted kernel based color histogram to compute the feature of object template and candidate regions, we proposed a robust Mean-Shift object tracking algorithm based on improved weighted background to enhance histogram saliency.The saliencies of the object and of the background were calculated out from their bin of histogram, which was incorporated into the traditional similarity measurement for defining an improved weighted background coefficient.Therefore, the discriminabiltity of the object from the background was increased, and the effect of background information on object locating was reduced.Experiment result shows that the tracking effect is improved apparently.