Development of a thermal-visible video surveillance system based on fractional order tv-model

Abstract

Abstract Detection and tracking of moving objects in video are an active and challenging research area of computer vision devoted to the applications of video surveillance. The several environmental conditions such as dark, foggy, snowing and rainy, and open-ended goals of the problem motivate to develop a robust surveillance system that based on the thermal-visible video spectrum fusion. However, visible-visible and thermal-thermal spectrum based surveillance model show insufficiency under such conditions. The detection of moving objects is performed by means of optical flow. This paper presents a novel fractional order total variation (TV) model in the estimation of optical flow. In particular, the presented fractional order TV-model is designed by generalizing an integer order total variation model formed by using a combination of total variation and quadratic regularization terms. However, it is challenging to solve such complex minimization problem due to the non-differentiability nature of fractional order TV-term. The fractional order derivative is discretized using the Grünwald-Letnikov derivative. The primal-dual algorithm is used as an iterative scheme to solve the resulting formulation. Finally, a number of experimental results on a fusion of visible-visible, thermal-thermal and visible-thermal video spectrum demonstrate the effectiveness of the model.