Multi-grid Methods For Variational Optical Flow Computations

People are paying more and more attention to the method of optical flow field in the past years. At the same time, more and more researchers apply this technology in the field of the video compression and the tracking of the motion object. In addition, the consistent development of the computer vision creates capacious development space for the optical flow field technology.There are several methods for computing optical flow. For example, variational methods, region-based matching methods, phase-based techniques, energy-based methods etc. Variational methods are widely used for computing optical flow. They can generally be divided into global methods and local methods. Global methods can yield 100% dense optical flow fields, while most local methods can only yield sparse optical flow fields. However they are more robust under noise. In this paper, at the base of combining local and global methods, increase a constraint in which gradient are changed, So as to achieve better results estimates. At the same time detailed description of the multi-grid methods, Use of the advantages of its rapid solution, solving the linear equations derived by optical flow calculation model. Compared with the traditional methods, multi-grid methods greatly reducing the iterative calculation. Also significantly reduce the time required, achieve the requirement of the real-time solution. In this paper, multi-grid method has been applied to solve P-M model and achieved more accurate solution than Gauss-Seidel iterative method.