基于广义关联聚类图的分层关联多目标跟踪

发布时间：2018-01-21 08:31

本文关键词： 多目标跟踪广义关联聚类图分层数据关联检测跟踪遮挡处理　出处：《自动化学报》2017年01期 　论文类型：期刊论文

【摘要】：检测跟踪是近期多目标跟踪研究的热点方向之一.目前大部分方法都是基于相邻帧之间的双向匹配,对检测点进行数据融合.本文提出的方法是,给定一个滑动时间窗口,在窗口内对某个目标每帧出现的检测点进行一次性数据融合.我们把多目标跟踪看作图的分割问题,利用广义关联聚类(Generalized correlation clustering problem,GCCP)图优化文中提出的数据融合.吸取分层数据关联的思想,把多目标跟踪分成两个阶段.首先,在时间窗口内遵循检测点,利用广义关联聚类,得到自适应长度的轨迹片段,轨迹片段长度不受窗口宽度的限制.然后,基于轨迹片段进一步数据关联,得到目标的长轨迹.在公共数据集上的实验测试表明,本文方法能够有效地实现多目标跟踪,对于遮挡处理、身份转换处理以及轨迹的生成具有很好的鲁棒性,多目标跟踪准确率(Multiple object tracking accuracy,MOTA)超过当前水平.
[Abstract]:Detection and tracking is one of the hot topics in the recent multi-target tracking research. At present, most of the methods are based on the bidirectional matching between adjacent frames and the data fusion of the detection points. The method proposed in this paper is. Given a sliding time window, the detection points of each frame of a target are fused in the window. We treat multi-target tracking as a graph segmentation problem. Generalized correlation clustering problem is used for generalized association clustering. GCCP-graph optimizes the data fusion proposed in this paper. By absorbing the idea of hierarchical data association, the multi-target tracking is divided into two stages. Firstly, the detection points are followed in the time window, and the generalized association clustering is used. The length of the trajectory segment is not limited by the window width. Then, the long trajectory of the target is obtained by further data association based on the trajectory fragment. The experimental results on the common data set show that the length of the trajectory fragment is not limited by the window width. This method can effectively achieve multi-target tracking, and has good robustness for occlusion processing, identity conversion processing and trajectory generation. Multiple object tracking acculturation rate (MOTAA) is higher than the current level.
【作者单位】：合肥工业大学计算机与信息学院;
【正文快照】： 引用格式齐美彬,岳周龙,疏坤,蒋建国.基于广义关联聚类图的分层关联多目标跟踪.自动化学报,2017,43(1):152-160Multi-object Tracking Using Hierarchical Data Association Based onGeneralized Correlation Clustering GraphsQI Mei-Bin1YUE Zhou-Long1SHU Kun1多目标跟踪是

【相似文献】