Resumen

One of the most interesting topics in the area of computer vision is the analysis of the movement of objects in a video sequence, a technique commonly known as optical flow. Although several methods have been proposed to solve the problem of optical flow, there is no optimal solution since it is a poorly planned problem, because the video images are only a two-dimensional projection of a 3D world. In this work a study of the original technique of Horn and Schunck is made that gave origin to the investigations in this area, and the use of multi-scale representations that have been used recently to improve the characteristics of the original technique. The multi-scale method is based on reducing the size of the image successively, thus a multi-level pyramid is constructed, starting from the original size or scale of each input image, in order to detect the smallest displacement, since that the original technique fails to detect movements larger than the size of a pixel. The algorithm used by this technique detects first the largest movements and the result is used as the starting point to detect the shortest movements, handling a pyramidal structure based on the previous procedure to calculate the movement at all the scales.