基于方向矢量的多特征融合粒子滤波人体跟踪算法研究  

Research of Human Tracking Algorithm through Multi Feature Fusion Particle Filter Based on Direction Vector

作　　者：张蕾[1] 宫宁生[1] 李金[1] 

机构地区：[1]南京工业大学电子与信息工程学院,南京211816

出　　处：《计算机科学》2015年第2期296-300,共5页Computer Science

基　　金：国家重点基础研究发展计划(973计划)(2005CB321901);软件开发环境国家重点实验室开放课题(BUAA-SKLSDE-09KF-03)资助

摘　　要：针对传统的多特征融合粒子滤波跟踪算法计算量大、不利于实时性、人群拥挤遮挡时容易出现跟踪匹配错误等情况,提出了基于方向矢量的多特征融合粒子滤波跟踪算法。该算法首先将人体颜色特征与轮廓特征进行乘性融合和加性融合后相加并加上两者的不确定性的乘积,以便能够根据两种特征的实际贡献率来调节各自在跟踪过程中所占的权重比例,从而提高了跟踪的准确性;其次结合方向矢量,根据先前的跟踪信息来预测运动物体可能运动的范围从而减少了粒子迭代计算量;最后通过动态调节窗口将合并的人体进行分离处理。实验证明,本方法能够在复杂情况下对人体进行实时准确的跟踪。Traditional multi feature fusion particle filter algorithm has huge amount of computation,which is not conducive to timeliness.Meanwhile,the algorithm often generates error for tracking and matching.To solve these problem better,this paper adopted the multi feature fusion particle filter algorithm founded in direction vector.Firstly,the algorithm adds the value of multiplicative fusion and additive fusion through the body color feature and contour feature.In addition,on the basis of the uncertainty multiplication of two features,the respective weights proportion in the tracking process is adjusted according to the actual contribution rate of two kinds of features.Moreover,the probable range of the movement of objects can be predicted to reduce the amount of calculation of particle iteration by combining the direction vector.At last,the merge human＇s body can be separated by adjusting the window automatically and precisely.The test shows our method can realize the human tracking accurately in the complex environment.

关 键 词：粒子滤波 多特征融合 人体跟踪 不确定性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]