Title | ||
---|---|---|
Robust Object Tracking via Structure Learning and Patch Refinement in Handling Occlusion. |
Abstract | ||
---|---|---|
Object tracking is a challenging task especially when occlusion occurs. In this paper, we propose a robust tracking method via structure learning and patch refinement to handle occlusion problem. First, we pose the tracking task as a structured output learning problem to mitigate the gap between pattern classification and the objective of object tracking. Contrary to the random target candidates selection method, we utilize the object independent proposal strategy to generate high quality training and testing samples in structured learning. Second, we over-segment the tracked target to a set of superpixel patches, and then train a background/foreground binary classifier to remove the background patches within the tracked object rectangle area for refining the tracking precision. The objective of target refining is to mitigate tracking model degradation and enhance model robustness for adapting our tracker for long-term and accurate tracking. Experimental results conducted on publicly available tracking dataset demonstrate that the proposed tracking method achieves excellent performance in handling target occlusion. |
Year | Venue | Field |
---|---|---|
2017 | ICIRA | Computer vision,Occlusion,Binary classification,Pattern recognition,Structure learning,Structured prediction,Rectangle,Active appearance model,Robustness (computer science),Video tracking,Artificial intelligence,Engineering |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
13 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Junwei Li | 1 | 3 | 2.77 |
Xiaolong Zhou | 2 | 103 | 19.67 |
Sheng-Yong Chen | 3 | 1077 | 114.06 |
Sixian Chan | 4 | 12 | 7.69 |
Zhaojie Ju | 5 | 284 | 48.23 |