Visual Tracking Based on Adaptive Spatially Regularized Correlation Filter

In this paper, a method for visual tracking based on an adaptive spatially regularized correlation filter is proposed. A correlation filter is learned from the training sample under the framework ridge regression. To alleviate the boundary effect caused by the assumption of a periodic extension of the training samples, a regularization weight matrix is devised in the form of a quadric function and then imposed on the correlation filter. The change between the two adjacent frames is measured by the correlation filter learned in the previous frame. The online adjustment of a parameter of the quadric function enables the correlation filter to adapt to the change in the appearance of the target. Finally, the target is determined among a collection of candidates by the learned correlation filter. The experiments on 100 video clips demonstrate the enhancement of the proposed method in the accuracy of tracking.