Learning Temporal Regularized Correlation Filter Tracker with Spatial Reliable Constraint

Abstract

Correlation filters have achieved appealing performance with high speed in recent years. The advantage of correlation filter-based tracking methods is mainly attributed to powerful features and effective online filter learning. However, the periodic assumption of the training data would introduce unwanted boundary effects, which severely degrade the discrimination power of the correlation filter. In this paper, we construct the spatial reliable map with deep features from Convolutional Neural Network, then the map is used to adjust the filter support to the part of the object suitable for tracking. In order to further improve the long-term tracking ability, we introduce temporal regularization to DCF training, which can deal with occlusion and deformation situations. The experimental results show that the proposed algorithm achieves high tracking success rate and accuracy.