Object tracking via null-space discriminative projections and sparse representation

The traditional target tracking algorithm based on sparse representation only considers the whole information of the target template without considering the information of the background. Tracking drift is easily happened when the target is disturbed by cluttered background, occlusion and illumination. Aiming at the existing problems, this paper proposes a sparse representation target tracking method based on null-space discriminative projection. On the one hand, the model increases the reconstruction error of the target sample by introducing the null-space discriminative projection method, thus improving the discriminative ability of the algorithm to the target and the background; On the other hand, using the L1 norm as the loss function reduces the sensitivity of the template to the outlier data. In addition, the model designs an online learning algorithm using to update the target tracking template. The tracking algorithm performs the best in the scene with high similarity between target and background. It can also deal with occlusion, illumination changes and other issues. The experimental results show that the proposed method is more stable, reliable and robust than the popular tracking algorithms. The specific experimental results are demonstrated in this paper.