Variational-Based Contour Tracking in Infrared Imagery

In this paper we address the problem of recovering object contour in infrared video sequences using active contours and level set methods. We propose an approach for variational segmentation of infrared images containing non-rigid, moving objects. The local regions-of-interest (ROIs) are identified firstly using statistical background-subtraction method. While the initialization of background model is not restricted to moving objects-free data. Within each ROI, contour curve evolves towards the direction of minimizing energy functional which dedicated to infrared spectrum imagery. The stopping function does not only account for boundary gradients, but also temporal coherence of object area and region signature of intensity distribution. The signature is optimal for discriminating foreground and background region. Experimental results for OTCBVS thermal dataset videos show the effectiveness and robustness of our method.