A Non-Local Weighted Fractional Order Variational Model for Motion Estimation

Motion estimation has been popular in vision system due to its many implementations, including video surveillance, weather forecasting, submarine navigation, object detection, fire detection, etc. Thus, effectively estimating motion in visual data has driven motion estimation research. Generally, motion estimation is conducted using optical flow, that is visualized by the vector plot and the color map flow field. The primary aim of this study is to incorporate the global and the local optical flow techniques in order to provide a robust and dense optical flow that preserves discontinuities. For this purpose, we proposed NF-VOF model called the nonlocal weighted variational model based on fractional derivative for optical flow estimation. The global nature of NF-VOF model encompasses the comprehensive motion information within the scene, while its local character pertains to the specific motion of every entity in the image frame. The combination of global-local nature along with Marchaud derivative in the model, successfully handles the edge discontinuities and textures. It also offers notable resistance against outliers. In order to conduct the thorough evaluation of NF-VOF model, variety of datasets are considered. Moreover, the efficiency of proposed model is shown against state-of-the-art techniques.