Abstract—Occlusion is a common problem encountered in
various tracking applications. This paper addresses occlusion
within the context of real-time tracking. Contributions of the
paper are two-fold. Firstly, the paper studies the occlusion
problem within the context of tracking-by-detection. Secondly,
a 3D approach that allows for tracking-by-detection algorithms
to be extended towards effective occlusion handling is put
forward. The proposed approach achieves an efficient
incorporation of depth features into the circulant tracking
framework, thereby achieving rapid object detection, tracking
and robustification. Finally, a patch-based modeling strategy
for depth features, coupled with a robust occlusion estimator is
proposed. The resulting scheme allows for the tracker to
achieve occlusion detection, tracker recovery and a significant
alleviation of the drift problem associated with
tracking-by-detection state-of-the-art. Experimental results on
benchmark sequences demonstrate the effectiveness and
robustness of the proposed scheme. The superiority of the
scheme is further established in comparison experiments with
state-of-the-art.
Index Terms—Circulant tracking, occlusion-handling, robust
tracking, 3D tracking.
Yao Yeboah is with the Department of Electrical and Computer
Engineering, South China University of Technology, Tianhe, Guangzhou,
China (e-mail: yeboahjunior@yahoo.com).
Zhuliang Yu and Wei Wu are with the School of Automation Technology,
South China University of Technology, Tianhe, Guangzhou, China (e-mail:
zlyu@scut.edu.cn, auweiwu@scut.edu.cn).
[PDF]
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.
