Non-negative matrix factorization based illumination robust meanshift tracking
Gargi Phadke, Rajbabu Velmurugan, Shubham Dawande · 2017
Object tracking is a critical task in surveillance and activity analysis. One main issue in tracking is illumination variation. We propose a method which is robust to illumination by incorporating a feature that is less variant to illumination. The proposed feature is a reflectance histogram obtained using sparsity constrained non-negative matrix factorization (NMFsc). Using NMFsc, illumination and reflectance components are separated in each frame of the input video. The obtained reflectance histogram is used as feature vector in a meanshift framework for target tracking. We also use an adaptive target model to handle target appearance changes. Experimental results using standard benchmark videos show that the proposed scheme can lead to better tracking in challenging illumination scenarios, when compared to several existing algorithms.