基于层叠分类器的快速相关跟踪算法

提出一种基于层叠分类器的快速相关跟踪算法。首先利用目标模板色彩分布信息对原始图像数据进行变换，锐化匹配相似度函数峰值，增强算法在复杂环境下的稳定性；然后提出了用平均灰度差和Harr型特征构造层叠分类器，分层刻画目标模板与搜索窗口在统计特征和局部特征上的相似性，并采用积分图快速计算特征，从而大幅度减少在非最优匹配点上的计算量，且特征计算与模板大小无关。大量实验结果表明，该算法大大降低了相关跟踪的时间复杂度，具有跟踪稳定、实时性强等特点。目前，以该算法为核心的实时目标跟踪系统对图像大小为320×240的视频序列内任意尺寸目标的平均处理速度达到20帧/s。

Real-time correlation tracking algorithm based on cascade classifier

A real time correlation tracking algorithm was presented based on cascade classifier. Firstly， to improve the robustness of the method in complex background， the original input video sequence was transformed according to the color distribution of target template. Then the average differences of gray value and Harr like features were designed to build a cascade classifier. The features in different levels represented the similarities between target template and searching window in various aspects like intensity distribution and gradient changes， so as to make full use of template information and greatly decrease the computational cost in false correlation position. Moreover， all features above were computed by integral image method. A real time visual tracking system was developed and tested under various conditions. Extensive experiment results demonstrate that the proposed algorithm is effective， and the processing speed of the system reaches 20 fps for the image size of 320×240.