基于多级跟踪队列的运动目标跟踪遮挡处理  

Occlusion Handling Method for Multiple Moving Objects Tracking Based on Multilevel Tracker Queues

作　　者：金标[1,2,3] 胡文龙[1,2] 王宏琦[1,2] 

机构地区：[1]中国科学院电子学研究所,北京100190 [2]中国科学院空间信息处理与应用系统技术重点实验室,北京100190 [3]中国科学院研究生院,北京100049

出　　处：《光学学报》2011年第8期211-218,共8页Acta Optica Sinica

基　　金：国家973计划(2010CB327900);国家自然科学基金(61001176)资助课题

摘　　要：针对多目标跟踪过程中存在的遮挡问题,提出了一种固定摄像机场景下的多目标实时跟踪算法。提出基于鬼影判别与背景模型选择更新的背景差法检测运动目标,建立一种融合色度与边缘特征的目标模型。通过定义稳定跟踪队列、临时跟踪队列、跟踪丢失队列以及候选跟踪队列等跟踪器队列,提出基于多级关联匹配的策略实现多目标跟踪遮挡处理,针对新目标、目标合并以及目标消失分别提出判别及跟踪策略。实验结果表明,运动目标检测方法能够抑制鬼影,防止缓慢运动的目标融入背景;同时,验证了目标模型的稳健性,以及跟踪算法能够在遮挡、交错等复杂情形下有效地跟踪多目标。A novel real-time multiple objects tracking algorithm is proposed to handle the problem of occlusion in the stationary situation. Moving objects are detected using the background subtraction based on ghost detection and selective background updating model, a stable object model fusing the hue and edge features is established, and multilevel tracker queues are defined to solve the occlusion, including stable tracker queue, temporal tracker queue, lost tracker queue, and uncertain tracker queue. Then the algorithm achieves tracking multiple objects handling occlusion based on multilevel data association, and it solves the problems of new objects appearing, objects merging, and object disappearing based on different strategies. Experimental results show that our detection method could restrain ghosts and prevent the moving objects from being fused to the background. Also, the method could testify the robustness of the object model and the tracking method could effectively track the objects in the complicated situation.

关 键 词：机器视觉 多目标跟踪 遮挡处理 多级关联跟踪 鬼影抑制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]