- 期刊
A High‐confidence Model Update Method for Kernel Correlation Filter Trackers
摘要
The correlation filter trackers have limited ability to represent the appearance model of the target, cannot effectively describe the appearance change. Besides, the model update strategy is unreliable and easily leads to model drift. To solve the above problems, we propose a high‐confidence model update method for kernel correlation filter trackers. First, we assign different weights to each feature and perform weighted feature fusion to improve the overall tracking performance. Secondly, in the model update stage, we propose a reliable model update strategy that can avoid the problem of model occlusion during the model update process. We conduct comparative experiments on the OTB‐ 2013 and OTB‐2015 datasets. The experimental results prove that the tracking performance of this method is better than other trackers. It has strong robustness to deformation and occlusion, and can effectively improve processing efficiency.