Correlation filter object tracking algorithm based on spatial and channel attention mechanism

Correlation filter algorithms are widely used in the field of object tracking due to their excellent tracking performance and real-time tracking efficiency. Traditional correlation filter tracking methods focus on three aspects of improving algorithm performance: feature representation, spatial regularization techniques, and temporal smoothness. However, these methods overlook the removal of redundant and interfering information in the filters, as well as the protection of filters in occluded scenes, which leads to poor performance in complex scenarios such as cluttered backgrounds and occlusions. It is noted that interfering information exhibits sparsity but often lacks the structural property of combinatorial sparsity. Inspired by this, this paper proposes a correlation filter object tracking algorithm based on spatial and channel attention mechanisms. Specifically, the algorithm introduces fiber group sparsity constraints in the spatial direction of the filter and slice group sparsity constraints in the row, column, and channel directions. In this way, the structural sparsity property of the filter is further exploited to allocate attention to the spatial and channel features, thereby removing redundant and interfering information. In addition, the reliability analysis of the best candidate samples is performed by a history template pool to decide whether the filter should be updated or not to avoid the tracking failure problem caused by filter degradation in occlusion scenarios. Experiments on several datasets show that the proposed method outperforms other state-of-the-art trackers and achieves good performance in terms of accuracy and robustness.