Robust visual tracking using self-adaptive strategy

Abstract

Listen

Translate article

Discriminative correlation filter-based algorithms have recently demonstrated prominent advantages in the community of computer visual tracking, due to their ability to convert ridge regression problems in the frequency domain for creating solutions efficiently, which has attracted a great deal of attention and spurred new research. High precision and robustness have always been the goals of visual tracking. However, during the tracking process, target objects often encounter sophisticated scenarios such as fast motion and occlusion. During this period, erroneous tracking information will be generated and delivered to the next frame for updating; the information will seriously deteriorate the overall tracking model. To address the problem mentioned above, in this paper, we propose an accurate model self-adaptive update method based on a discriminative correlation filter framework. The proposed tracking method is achieved by utilizing the peak score of a response map generated by the discriminative correlation filter as a dynamic threshold with comparisons to its PSR (peak side-lobe ratio) scores, and then the comparative results are used as the differentiated condition for updating the translation filter and scale filter model. In addition, multiple hand-crafted features such as HOG (histogram of oriented gradient), CN (color names), and HOI (histogram of local intensities) are fused self-adaptively for comprehensive feature representation, which further improve tracking performance. We evaluate the performance of the proposed tracker on OTB benchmark datasets; the experimental results demonstrate that the proposed tracker performs favorably against most state-of-the-art discriminative correlation filter-based trackers including some methods follow deep learning paradigm, and the effectiveness of updating the model self-adaptive is verified.