Robust visual tracking via incremental low-rank features learning

In this paper, we address robust visual tracking as an incremental low-rank features learning problem in a particle filter framework. Our new algorithm first learns the observation model by extracting low-rank features and the corresponding subspace basis of the object from the initial several frames. Then the low-rank features and sparse errors can be incrementally updated using an ℓ1 norm minimization model. We show that the proposed strategy is actually an online extension of Robust PCA (RPCA). Thus compared with previous methods, which directly learn subspace from corrupted observations, our model can incrementally pursuit the low-rank features for the target and detect the occlusions by the sparse errors. Furthermore, the proposed reformulation of RPCA can also be considered as an illumination study on extending batch-mode low-rank techniques for more general online time series analysis tasks. Experimental results on various challenging videos validate the superiority over other state-of-the-art methods.