Optical Flow Based on Lucas-Kanade Method for Motion Estimation

Abstract

Optical flow is considered an important step in many computer vision applications. Some optical flow estimation approaches are mainly based on correlation, gradient and frequency information respectively. Among algorithms most known for computing Optical Flow vectors are Lucas-Kanade and Horn-Schunck. In this paper, Optical Flow Based Lucas-Kanade method is implemented for detecting moving objects. First, optical flow computation between consecutive frames of an image sequence is calculated. Then, the resultant optical flow vectors obtained is coded to color to have a visualized motion estimation. Finally, a segmentation step is performed on the resulted motion estimation to highlight objects of interest. Simulation results on some image sequences show that the estimated flow accurates the motion for each and every pixel. A quantitive evaluation is performed on the Middlebury dataset for optical flow. Results obtained on the Middleburry dataset show that the estimated optical flow is promising and can be applied as a preliminary task to many treatments higher level in computer vision such as recognition or tracking.