Spatial and temporal constraints are very important for correlation filter (CF) based trackers. However, existing methods usually fail to regularize the CF learning by the spatio-temporal information because of the ineffective target representation. To address the issue, we propose a novel Orientation and Scale adaptive tracker with Regularized Correlation Filters (OSRCF) for visual tracking. First, we directly employ the target region to build a spatio-temporally regularized CF, which is solved efficiently by the alternating direction method of multipliers (ADMM). Specially, the spatial regularization module can alleviate boundary effects by suppressing the outside background, while the temporal one can handle rapid appearance changes by smoothing the CF updating. Second, we obtain a non-axis-aligned bounding box for the target region representation by an orientation and scale adaptive strategy, which cooperates a straightforward orientation estimation with a discriminative scale space correlation filter. We perform comprehensive experiments on two recent visual tracking benchmark datasets: VOT2017 and OTB2015. Results show that our OSRCF tracker outperforms top-ranked methods with handcrafted features in VOT2017 and achieves outstanding performance compared to some state-of-the-art methods in OTB2015.

When the target is partially occluded or undergoes drastic appearance change, drift is prone to occur in traditional correlation filter-based trackers, and ultimately leads to tracking failure. This paper proposes a robust tracking algorithm which can be applied to practical engineering. The proposed algorithm divides all training samples into different training sets based on their similarities. An independent filter is trained for every training set. In each frame of the tracking process, the similarity between candidate region and any training set’s training samples is calculated to select the most matched filter for target location. As a newly introduced training sample, feature map within current frame’s estimated target bounding box is assigned to the most similar training set to update the most matched filter. The updating of each filter is relatively independent. Each filter corresponds to one kind of target’s typical appearance, so the proposed tracking algorithm can memorize a variety of typical target appearances that appeared in the past and adapt to target’s discontinuous appearance change. The detector improves tracking accuracy and assist occlusion judgment. After occlusion, the target may reappear with a different posture. As long as similar postures have been captured in the previous tracking process, the target can be retrieved accurately while not being confused with other objects of target’s category. We evaluated our method on massive videos shot in actual engineering sites, and the result demonstrates the proposed tracking algorithm can handle occlusion and target’s posture changes very well and significantly outperforms other tracking algorithms.