RGBT tracking based on cooperative low-rank graph model

The existing graph-based RGBT tracking methods mainly focus on assigning a weight to each local image patch to suppress background influence in target bounding box, but the influences of background clutter might limit the improvement of tracking performance. To solve this problem, we propose a new algorithm, called cooperative low-rank graph model, to suppress background clutter. Specifically, the proposed feature decomposition module decomposes input dual-modal features into low-rank components and sparse noisy components, which could be used collaboratively by regularizing graph learning by combining modal weights. Besides, to avoid SVD (Singular Value Decomposition) operations we have designed an efficient solver based on ADMM (Alternating Direction Methods of Multipliers), which could factorize the low-rank matrix into two low-dimensional submatrices. Extensive experiments on four RGBT tracking benchmark data sets show that our method performs favorably against other state-of-the-art tracking algorithms, and achieves more robust tracking performance.