Enable Scale and Aspect Ratio Adaptability in Visual Tracking with Detection Proposals

Dafei Huang, Lei Luo, Mei Wen, Zhaoyun Chen and Chunyuan Zhang

Abstract

The newly proposed correlation filter based trackers can achieve appealing performance despite their great simplicity and superior speed. However, this kind of object trackers is not born with scale and aspect ratio adaptability. To tackle this problem, this paper integrates the class-agnostic detection proposal method, which is widely adopted in object detection area, into a correlation filter tracker. Additional optimizations such as feature integration and proposal rejection are also applied to make detection proposals more helpful. To reveal the effectiveness of our approach, two experiments are performed on 28 benchmark sequences with significant scale variation and 14 sequences with obvious aspect ratio change respectively. Among state-of-the-art trackers and existing scale-adaptive correlation filter variants, our proposed tracker reports the highest accuracy. Best accuracy is also achieved on the whole 50-sequence dataset with various challenging attributes at an average speed of 20.8 FPS, which proves the robustness and efficiency of our tracker.

Session

Tracking and Pose Estimation

Files

Extended Abstract (PDF, 571K)
Paper (PDF, 1900K)

DOI

10.5244/C.29.185
https://dx.doi.org/10.5244/C.29.185

Citation

Bibtex

@inproceedings{BMVC2015_185,
	title={Enable Scale and Aspect Ratio Adaptability in Visual Tracking with Detection Proposals},
	author={Dafei Huang and Lei Luo and Mei Wen and Zhaoyun Chen and Chunyuan Zhang},
	year={2015},
	month={September},
	pages={185.1-185.12},
	articleno={185},
	numpages={12},
	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
	publisher={BMVA Press},
	editor={Xianghua Xie, Mark W. Jones, and Gary K. L. Tam},
	doi={10.5244/C.29.185},
	isbn={1-901725-53-7},
	url={https://dx.doi.org/10.5244/C.29.185}
}