IEEE Transactions on Pattern Analysis and Machine Intelligence
Incremental Learning for Robust Visual Tracking
International Journal of Computer Vision
Online visual tracking with histograms and articulating blocks
Computer Vision and Image Understanding
ICPR '10 Proceedings of the 2010 20th International Conference on Pattern Recognition
On-line inverse multiple instance boosting for classifier grids
Pattern Recognition Letters
Online discriminative object tracking with local sparse representation
WACV '12 Proceedings of the 2012 IEEE Workshop on the Applications of Computer Vision
ICCV '11 Proceedings of the 2011 International Conference on Computer Vision
Visual tracking with representative templates based on low-rank matrix
Proceedings of the 2013 Research in Adaptive and Convergent Systems
| Hi-index | 0.00 |
Almost all the online tracking methods suffer from the problem of drift. The critical reason lies in that they only focus on how to get the most adaptive visual model to the last state(s), but lose the memory of the learned features. Aiming to address this challenging problem, we propose a novel appearance model within the framework of Bayesian-based tracking. The appearance model combines with the most representative basic models learned in the past temporal space. These basic representative appearance models, as the accumulated memory of the different aspects of features from different learning periods, can approximate the target appearance of the coming state as well as possible. Meanwhile, we design methods to online construct and update the basic models. Also, a dynamic scheme is employed to integrate the basic models. The novel model proposed in this paper, by explicit inference, can effectively and efficiently handle the challenging cases such as full occlusion, frequent and drastic appearance variation of the target. As well, it is demonstrated by the extensive experiments.