High speed long-term visual object tracking algorithm for real robot systems

Muxi Jiang,Rui Li,Qisheng Liu,Yingjing Shi,Esteban Tlelo-Cuautle
DOI: https://doi.org/10.1016/j.neucom.2020.12.113
IF: 6
2021-04-01
Neurocomputing
Abstract:<p>Although many visual tracking algorithms have made many achievements in video sequences, they have not been confirmed to work well on the real robot systems with the unpredictable changes and limited computing capabilities. In order to face the complex practical conditions, including huge scale variation, occlusion and long-term task, this paper develops a CF-based long-term tracking algorithm. The main strategies are as follows. A novel confidence score is proposed to judge tracking reliability, and the tracking drift is corrected to keep the target's long-term appearance. Furthermore, once the target is lost, it can be relocated by the multi-scale search. Our tracker performs favorably against other CF-based trackers with strong engineering applicability. Finally, experiments on the datasets and an UAV are carried out to verify the effectiveness for real robot systems.</p>
computer science, artificial intelligence
What problem does this paper attempt to address?