计算机科学与探索 ›› 2020, Vol. 14 ›› Issue (6): 1036-1044.DOI: 10.3778/j.issn.1673-9418.1904011

• 图形图像 • 上一篇    下一篇

轨迹树层次关系模型多摄像机多目标跟踪研究

刘冠群,李婷   

  1. 1. 湖南广播电视大学 网络资源系,长沙 410004
    2. 中南大学 信息科学与工程学院,长沙 410083
  • 出版日期:2020-06-01 发布日期:2020-06-04

Research on Multi-camera Multi-target Tracking Method Based on Hierarchical Relational Model of Trajectory Tree

LIU Guanqun, LI Ting   

  1. 1. Department of Network Resources,  Hunan  Radio & TV  University, Changsha 410004, China
    2. College of Information Science and Engineering, Central South University, Changsha 410083, China
  • Online:2020-06-01 Published:2020-06-04

摘要:

为提高摄像机目标跟踪精度,提出基于多假设跟踪(MHT)框架的采用轨迹树层次关系模型多摄像机多目标跟踪方法。首先,通过多个摄像机产生的轨迹之间的时空关联,找出未知数目的多个轨迹,并通过求解各帧的最大加权问题(MWCP),在线实现对目标三维轨迹的估计。其次,为解决多帧图像处理的NP难问题,提出了一种新的在线方案,该方案利用前一帧结果的反馈信息,建立多帧图像处理方案,从而在每一帧上找到最佳的轨迹。该方案能使多个子问题构成多目标多目标控制问题,大大减少了计算量。实验表明,该算法与目前最先进的批处理算法相比,具有较好的性能。

关键词: 轨迹树, 层次关系, 多摄像机, 多目标, 跟踪, 多帧图像

Abstract:

To improve the accuracy of camera target tracking, a multi-camera multi-target tracking method based on multi-hypothesis tracking (MHT) framework using hierarchical relation model of trajectory tree is proposed. First of all, the unknown number of trajectories can be found by spatiotemporal correlation between trajectories generated by multiple cameras, and the three-dimensional trajectory of the target can be estimated online by solving the maximum weight clique problem (MWCP) of each frame. Then, in order to solve the NP problem of multi-frame image processing, a new online scheme is proposed, which uses the feedback information of the results of the previous frame to establish a multi-frame image processing scheme, so as to find the best trajectory in each frame. This scheme can make multiple sub-problems constitute multi-objective and multi-objective control problems, and greatly reduces the amount of calculation. Experiments show that the algorithm has better performance than the most advanced batch processing algorithm.

Key words: trajectory tree, hierarchical relationship, multi-camera, multi-target, tracking, multi-frame image