Illumination Invariant L1 Tracker Using Photometric Normalization Techniques

摘要

Recently, sparse representation-based tracking methods called l1 trackers give remarkable performances in difficult video sequences. However, the tracking in the situation of large illumination changes and shadow casting still has serious problems that need to be solved. A new illumination invariant tracking method based on photometric normalization techniques and sparse representation framework is proposed. By using photometric normalization methods, we create a new illumination invariant template presentation for tracking and eliminate the effect of brightness variation and shadow casting. For enhancing the tracking accuracy, a method for adaptively selecting the optimal template presentation at the update step of the tracking process is introduced. The experiments show that our method outperforms the previous l1 tracker and some state-of-the-art tracking algorithms in challenging tracking sequences.