Joint Conditional Random Fields for Multi-object Tracking with a Mobile Robot

A novel joint conditional random field(JCRF)with hierarchical structure is proposed for multi-object tracking of mobile robots by abstracting the data association between objects and observed data to be a sequence of labels.JCRF include two layers of random fields,one for joint data association and the other for moving object state estimation.With JCRF,shape information and motion information can be fused for joint data association to improve the stability of object tracking,and moving object detection and object tracking can be performed simultaneously.In this paper,multi-object tracking based on laser range finder is studied using JCRF in which candidate regions of object are segmented out from laser range sensor data firstly and then the match tree is adopted to reduce the state space of label sequence.Experimental results on the mobile robot show that the multi-object tracking based on JCRF can run precisely and stably in real time.