This paper researched the problem that the traditional Mean-Shift algorithm could not track a size-changing object effectively,and proposed a new algorithm based on assistant decision-making of object similarity metric,to estimate the scale and orientation of a tracking window with adaptive bandwidth.Firstly,it adopted the saliency of object and background to im-prove tracking accuracy for orientation,and then employed the local exhaustive search to compute the similarity metric between object model and the certain region where was around the tracking center in each frame.Finally,it determined the object scale variation by similar pixel amounts.What’s more,it defined a novel bandwidth criterion for improving adaptability in tracking bandwidth.The experimental results prove that the present method can improve the tracking accuracy effectively in orientation between space and scale.%针对传统窗宽固定不变的Mean-Shift跟踪算法不能实时地适应目标尺寸大小变化这一问题，提出了一种基于目标相似度辅助决策的带宽自适应跟踪算法。首先利用目标与背景的特征显著性，提高跟踪算法空间定位准确性；然后利用局部穷搜索的方法，计算目标模型与每一帧目标跟踪中心点附近一定区域的相似性；最后通过统计分析前后帧相似像素点数目变化，确定目标尺度变化情况，从而建立一种自适应更新带宽准则，提高算法对目标尺度变化的自适应性。实验结果表明，改进的算法可以有效地提高Mean-Shift跟踪算法空间和尺度定位准确性。
展开▼
