改进Camshift算法的多行人目标跟踪方法  

Multi-pedestrian target tracking method based on improved Camshift algorithm

作　　者：陈艺[1] CHEN Yi(Sichuan University of Arts and Sciences,DaZhou,SiChuan635000,China)

机构地区：[1]四川文理学院

出　　处：《导航定位学报》2019年第4期30-36,共7页Journal of Navigation and Positioning

基　　金：四川省教育厅科研项目(17ZB0376)

摘　　要：针对基于颜色特征的目标跟踪方法在跟踪多个行人目标时,易受衣服颜色相近的行人影响,造成行人目标跟踪发生错误的问题,提出一种改进Camshift算法的多行人目标跟踪方法:为克服单一颜色特征作为目标模型易造成目标丢失的不足,按一定的权值系数融合目标的颜色特征和HOG特征来建立目标模型;并分别对多个行人目标建立目标模型,将传统的Camshift算法的单目标跟踪扩展成多目标跟踪。实验结果表明,该方法相比于传统Camshift算法更具鲁棒性,跟踪准确率可提升5.3%,相比于粒子滤波算法,实时性能够提升30.23%。Aiming at the problem that it is susceptible to multi-pedestrian targets with similar clothing colors for the color-based target tracking method,which results in the error in tracking targets,the paper proposed a multi-pedestrian target tracking method with the improved Camshift algorithm:in order to overcome the shortcoming that the target is easily lost with single color feature as target model,the target model was established by integrating a certain weight coefficients with the color and HOG features of the target;and the models for multiple pedestrian targets were built for extending the single target tracking of traditional Camshift algorithm to multi-target tracking.Experimental result showed that the proposed method would be more robust,and the tracking accuracy could be improved by 5.3%compared with traditional Camshift algorithm;meawhile,the real-time performance could be improved by 30.23%compared with particle filter algorithm.

关 键 词：CAMSHIFT算法 颜色特征 多目标跟踪:融合系数 

分 类 号：P228[天文地球—大地测量学与测量工程]