视频序列中面向行人的多目标跟踪算法  

Pedestrian Oriented Multi-Object Tracking Algorithm in Video Sequence

作　　者：李琦[1] 邵春福[1] 岳昊[1] 

机构地区：[1]北京交通大学城市交通复杂系统理论与技术教育部重点实验室,北京100044

出　　处：《北京理工大学学报》2013年第2期178-184,共7页Transactions of Beijing Institute of Technology

基　　金：国家重点基础研究发展计划资助课题(2012CB725403)

摘　　要：为实现视频序列中多行人目标跟踪,基于多信息融合方法,考虑多目标间严重遮挡,建立面向行人的多目标跟踪算法.提出多信息融合算法融合目标颜色和运动信息,结合均值漂移算法思想,实现常态下目标跟踪.针对多行人目标参与的遮挡,通过理论分析遮挡过程中目标面积变化,提出遮挡因子判别遮挡发生、辨识遮挡者和被遮挡对象、确认被遮挡对象重新出现等.实验结果表明,该方法能够正确跟踪行人目标,判断并处理多目标间的严重遮挡.To track pedestrian oriented multiple objects in video sequence, this paper presents a tracking algorithm using multi-cue integration approach and high degree occlusion processing. Combined with the mean-shift algorithm, multi-cue integration method could track object in normal condition by fusing the color and motion information. To deal with the issue of occlusion involving multiple objects, the area change of the object during occlusion has been analyzed in theory. An occlusion factor is advanced to detect occlusion occur, identify obstructer or occluded object and to check the reappearance of an occluded object. Experiment results show that the proposed tracking algorithm could correctly find out and resolve the occlusion among the multiple objects so that it could successfully track the multiple pedestrians.

关 键 词：智能交通 多目标跟踪 行人跟踪 多信息融合 严重遮挡 

分 类 号：U491[交通运输工程—交通运输规划与管理] TP391[交通运输工程—道路与铁道工程]