doi: 10.18178/ijfcc.2017.6.1.480
Towards Occlusion Handling in Visual Tracking-by-Detection Systems
Yao Yeboah
1
,
Zhuliang Yu
2
,
Wei Wu
2
- 1 Department of Electrical and Computer Engineering, South China University of Technology, Tianhe, Guangzhou, China.
- 2 School of Automation Technology, South China University of Technology, Tianhe, Guangzhou, China.
Abstract
Occlusion is a common problem encountered invarious tracking applications. This paper addresses occlusionwithin the context of real-time tracking. Contributions of thepaper are two-fold. Firstly, the paper studies the occlusionproblem within the context of tracking-by-detection. Secondly,a 3D approach that allows for tracking-by-detection algorithmsto be extended towards effective occlusion handling is putforward. The proposed approach achieves an efficientincorporation of depth features into the circulant trackingframework, thereby achieving rapid object detection, trackingand robustification. Finally, a patch-based modeling strategyfor depth features, coupled with a robust occlusion estimator isproposed. The resulting scheme allows for the tracker toachieve occlusion detection, tracker recovery and a significantalleviation of the drift problem associated withtracking-by-detection state-of-the-art. Experimental results onbenchmark sequences demonstrate the effectiveness androbustness of the proposed scheme. The superiority of thescheme is further established in comparison experiments withstate-of-the-art.
Keywords
- Circulant tracking
- occlusion-handling
- robusttracking
- 3D tracking
How to Cite
Yao Yeboah, Zhuliang Yu, and Wei Wu, "Towards Occlusion Handling in Visual Tracking-by-Detection Systems," International Journal of Future Computer and Communication, vol. 6, no. 1, pp. 6-14, 2017. https://doi.org/ 10.18178/ijfcc.2017.6.1.480
Copyright & License
Copyright © 2017 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).