Multi Feature Region Descriptor Based Active Contour Model for Person Tracking

In this paper, we propose a new region based active contour method for person tracking. The method combines multiple cues such as color, texture and shape information to track the objects. The extracted features are enrolled in a covariance matrix which captures not only each feature variation but also their correlations. The tracking is formulated by minimizing an energy functional using level-set method. The proposed method is robust to illumination, appearance changes, deformations, scale variations and occultation. Experimental results approve its efficiency and accuracy for many applications such as smart home monitoring and video surveillance.