Video surveillance is known as a fundamental task in computer vision, which aims to detect and track the moving objects in an image sequence (video). For this purpose, visible-visible spectrum system have been frequently used for various applications. However, it is not feasible for adverse environmental conditions such darkness, fog, snowfall and storms, etc. Thus, in order to address these challenges, this paper integrates the visible (color) and infrared (thermal) spectrum to provide a robust surveillance system. The object motion detection is performed in terms of optical flow, which is estimated using the Geman-McClure norm based fractional order variational model. This variational function is convex in nature and robust against outliers, and provides a dense flow field. The minimization of this variational function is done using Euler-Lagrange (EL) equations and numerically discretized based on the Grunwald-Letnikove (GL) derivative definition. The resulting non-linear system of equations is converted to a linear system using a fixed-point iteration scheme and finally solved by employing the efficient numerical scheme. The significance of the model is shown by performing the experiments on the fusion of image pairs, visible-thermal (VT), thermal-thermal (TT) and visible-visible (VV) spectrum.