Optical Flow Detection Algorithm and Application Based on Accelerated Function

Abstract

A novel approach that integrates multigrid computation and Harris corner extraction is proposed for solving the problems of high complexity and low speed in conventional optical flow computation. Firstly, Harris corners are extracted from grabbed frame images of real-time video. Then corner images are generated on the basis of corner coordinates. Secondly, an efficient computational technique based on multigrid is adopted to optimize the optical flow estimation on the consecutive corner images. Therefore, real-time movement can be detected by optical flow vector field. After many experiments on videos with moving objects in different backgrounds and rates of frame grabbing, experimental results demonstrate the higher computational speed. This approach is applied to detect moving objects in real-time videos and the results are very encouraging. The approach aims at accelerating optical flow computation, also being robust to disturbance of complex background due to only adopting corner images.