面向移动单摄像机的多目标跟踪算法  

Multi-object tracking algorithm for mobile single camera

作　　者：李兴玮[1] 陈慧敏 吕林珏 关少杰 LI Xingwei;CHEN Huimin;LYU Linjue;GUAN Shaojie(College of Intelligence Science and Technology, National University of Defense Technology, Changsha 410073, China)

机构地区：[1]国防科技大学智能科学学院,湖南长沙410073

出　　处：《国防科技大学学报》2020年第6期120-126,共7页Journal of National University of Defense Technology

摘　　要：针对移动单摄像机采集的视频序列中的运动多目标,重点研究了基于目标间的相对运动信息和数据关联策略的在线多目标自动跟踪器。利用目标间相对运动模型实现目标轨迹的恢复,减少目标轨迹碎片。运用事件匹配算法改进当前帧的检测响应与过去轨迹的分配,并降低跟踪过程中的目标身份转换次数。实验结果表明:该改进算法较原算法能够对序列中目标跟踪定位得更加精确,减少了轨迹碎片和身份转换次数,在TUD-Campus序列上达到了与国际前沿多目标跟踪算法相当的效果。Aiming at the moving objects in the video sequences collected by the mobile single camera,an online multi-object automatic tracker was focused on research,which was based on the relative motion information and data association strategies.The recovery of the object trajectory was achieved by the relative motion model between the objects,and the object trajectory fragmentation was reduced.The assignment between detection of the current frame and the past trajectories was improved based on the event matching algorithm,which reduced the number of identity conversions during tracking.Experimental results show that the improved algorithm is more accurate than the original algorithm in tracking and positioning the object in the sequences,reducing the number of trajectory fragmentation and identity conversion,and our tracker achieve relative the same performance in the state-of-the-art on the TUD-Campus sequence in the international academic circle.

关 键 词：相对运动模型 事件匹配算法 数据关联 移动单摄像机 多目标跟踪器 

分 类 号：TN95[电子电信—信号与信息处理]