Evaluation of Efficient Methods for Optical Flow Computation

Evaluation of Efficient Methods for Optical Flow Computation. The efficient computation of optical flow is a prerequisite for a large number of applications, which led to the development of various different optical flow algorithms in the past years. Optical flow describes for each pixel in the first image its position in the second image. Hence, it provides the basis to analyze ego-motion or motion of objects between two images in a pixel-wise fashion. Despite of the fact that many efficient solutions to calculate the optical flow have been developed in the past, there is a lack of research simultaneously comparing the accuracy and the run-time efficiency of these algorithms systematically. After the mathematical definition of the optical flow, the basic principles of four selected methods of optical flow computation are explained. The main contribution of this article is the evaluation of these algorithms. The studied algorithms are on the one hand so called multigrid methods and on the other hand approaches that are based on wavelet decomposition. All algorithms have in common that they prospect high performance. The motivation of the evaluation is to help the reader to find the most suitable algorithm for his task.