中文摘要：

针对目标跟踪中局部特征模型容易受到形变和误匹配等影响造成漂移的问题,提出了一种利用全局特征对霍夫局部模型进行约束的目标跟踪算法.该算法通过检测和提取目标的局部特征,构建了一个基于局部特征投票的概率模型.该模型以局部特征为基本元素,利用广义霍夫变换对局部特征的稳定性进行加权,通过对局部特征的增减维护实现模型的在线更新;将稳定的局部特征作为前景,计算目标全局颜色特征的概率分布,实现简单的柔性目标分割,用于调整局部特征模型并对跟踪进行优化,从而减少形变等问题产生的跟踪误差;局部特征和全局特征相互约束补充,共同完成在线自适应更新.实验结果表明,该算法有效提高了局部特征模型的准确率和稳定性,对部分遮挡和复杂形变的情况表现良好.

英文摘要：

A visual tracking method is proposed to solve the problem that the local feature model is prone to be influenced by the deformation and feature mismatch and leads to drifting. The method constrains the Hough local model by using global features. The Hough model is constructed with a set of local features of an object and estimates the stability of each local feature based on the generalized Hough Transform. Then, the model is updated flexibly through modifying the local feature set to adapt to the object appearance variations. Color cues of the local features are used to calculate the global color probability of being foreground or background. Then the global probability is used to adjust the tracking results and constrains the local features of the model in return. Experimental results on several public video sequences show the robustness of the proposed method in tracking objects with deformation and partial occlusion.