Extendable Multiple Nodes Recurrent Tracking Framework With RTU plus
Shuai Wang,Hao Sheng,Da Yang,Yang Zhang,Yubin Wu,Sizhe Wang
DOI: https://doi.org/10.1109/TIP.2022.3192706
IF: 10.6
2022-01-01
IEEE Transactions on Image Processing
Abstract:Recently, tracking-by-detection has become a popular paradigm in Multiple-object tracking (MOT) for its concise pipeline. Many current works first associate the detections to form track proposals and then score proposalns by manual functions to select the best. However, long-term tracking information is lost in this way due to detection failure or heavy occlusion. In this paper, the Extendable Multiple Nodes Tracking framework (EMNT) is introduced to model the association. Instead of detections, EMNT creates four basic types of nodes including correct, false, dummy and termination to generally model the tracking procedure. Further, we propose a General Recurrent Tracking Unit (RTU++) to score track proposals by capturing long-term information. In addition, we present an efficient generation method of simulated tracking data to overcome the dilemma of limited available data in MOT. The experiments show that our methods achieve state-of-the-art performance on MOT17, MOT20 and HiEve benchmarks. Meanwhile, RTU++ can be flexibly plugged into other trackers such as MHT, and bring significant improvements. The additional experiments on MOTS20 and CTMC-v1 also demonstrate the generalization ability of RTU++ trained by simulated data in various scenarios.