Abstract | ||
---|---|---|
Visual object tracking is a popular but challenging problem in computer vision. The main challenge is the lack of priori knowledge of the tracking target, which may be only supervised of a bounding box given in the first frame. Besides, the tracking suffers from many influences as scale variations, deformations, partial occlusions and motion blur, etc. To solve such a challenging problem, a suitable tracking framework is demanded to adopt different tracking scenes. This paper presents a novel approach for robust visual object tracking by multiple features fusion in the Siamese Network. Hand-crafted appearance features and CNN features are combined to mutually compensate for their shortages and enhance the advantages. The proposed network is processed as follows. Firstly, different features are extracted from the tracking frames. Secondly, the extracted features are employed via Correlation Filter respectively to learn corresponding templates, which are used to generate response maps respectively. And finally, the multiple response maps are fused to get a better response map, which can help to locate the target location more accurately. Comprehensive experiments are conducted on three benchmarks: Temple-Color, OTB50 and UAV123. Experimental results demonstrate that the proposed approach achieves state-of-the-art performance on these benchmarks. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1007/978-3-319-97304-3_58 | PRICAI 2018: TRENDS IN ARTIFICIAL INTELLIGENCE, PT I |
Keywords | Field | DocType |
Deep learning, Siamese Network, Object tracking, Feature fusion | Correlation filter,Pattern recognition,Computer science,Motion blur,Fusion,Video tracking,Eye tracking,Artificial intelligence,Deep learning,Economic shortage,Minimum bounding box | Conference |
Volume | ISSN | Citations |
11012 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 21 | 6 |
Name | Order | Citations | PageRank |
---|---|---|---|
Dongyan Guo | 1 | 24 | 6.49 |
Weixuan Zhao | 2 | 0 | 0.68 |
Ying Cui | 3 | 30 | 6.80 |
Zhenhua Wang | 4 | 12 | 3.23 |
Shengyong Chen | 5 | 8 | 6.22 |
Jian Zhang | 6 | 1305 | 100.05 |