一种基于高斯概率模型的多人跟踪算法.pdfVIP

Multiple Pedestrians Tracking based on Gaussian Probability Model DU Xin, CHEN Jian-juan, WANG Xuan-he (Department of Information Science Electronic Engineering, Zhejiang University, Zhejiang Hangzhou, 310027 ) Abstract: To deal with the pedestrians tracking problem in video surveillance, an algorithm based on Gaussian probability was proposed. Modeling object on the basis of statistical characteristics of color information which was grouped by improved k-means algorithm, this algorithm segmented object into blocks according to the grouping results, and utilized a Gaussian function in order to describe the distribution of color information in each block as well as the position information of pedestrians. The tracking algorithm accomplished when the maximum joint probability of color and position was obtained. Along with position information and color information of objects among adjacent frames, the failure case which is because of information loss due to occlusion could be solved. The experimental results validate the performance of the proposed algorithm. Moreover, they also show the robustness and capability of multiple-target tracking. Key words: Gaussian probability model, color clustering, multi-pedestrians tracking, occlusion 一种基于高斯概率模型的多人跟踪算法 杜歆，陈建娟，王选贺 （浙江大学信息与电子工程学系，浙江杭州，310027 ） 摘要：针对视频监控中的多人跟踪问题，提出了一种基于高斯概率模型的算法。基于目标颜色的统计特征， 采用改进的 K 均值方法，将目标区域按颜色信息聚类，并根据聚类结果对目标区域分块，然后用高斯模型 对各分块分别进行建模。同时，对目标的位置信息也进行高斯建模。通过计算待检测目标与模型之间颜色 和位置的最大联合概率值来实现跟踪。利用前后帧中目标的位置信息及颜色信息，能克服目标遮挡后因信 息的丢失而跟踪失败的情况。实验结果表明，该算法具有较强的鲁棒性，能有效实现多人的跟踪。 关键词：高斯概率模型；颜色聚类；多人跟踪；遮挡 中图分类号：TP39 文献标识码 :A 对人体目标的视频跟踪是计算机视觉 人个数，但当人头部不是在最外层轮廓时就 研究中的热点也是难点。学者们提出了许多 会出错。McKenna[5]根据颜色信息，用人体 方法：如基于特征的跟踪[1]，基于贝叶斯推 的颜色直方图来分割人群，但由于颜色信息 断的跟踪[2]等。其中基于特征的跟踪通常先