Feature Combination Tracking

Multiple kernel learning methods have been successfully applied to visual tracking by finding the best combination of the kernels using boosting techniques. However, they are still not effective in tracking objects with large appearance variations and are not able to generate qualified combinations....

Full description

Saved in:
Bibliographic Details
Main Authors: Feng, Shaokun, Zhao, Yanan
Format: Conference Proceeding
Language:English
Subjects:
Boosting
Computational modeling
Feature combination
Kernel
Multiple features
Target tracking
Tracking
Training
Visualization
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Multiple kernel learning methods have been successfully applied to visual tracking by finding the best combination of the kernels using boosting techniques. However, they are still not effective in tracking objects with large appearance variations and are not able to generate qualified combinations. In this paper, we propose a tracking framework by leveraging the multiple features via feature combination. By using multiple features combined with the LPBoost technique, our method is able to compute better combinations than others. This framework exploits both local and global features of image patches, thereby generating more accurate tracking results. In addition, a template update strategy is introduced to let the proposed framework more robust to partial occlusion. Experiments on challenging video sequences demonstrate that the proposed tracking algorithm outperforms several state-of-theart methods in both qualitative and quantitative evaluations.
ISSN:2473-3547
DOI:10.1109/ISCID.2017.101