Determining the optimal filter for the Lucas-Kanade optical flow algorithm

The work is devoted to determining of optimal filter for the Lucas-Kanade optical flow algorithm. The work shows description of Lucas-Kanade algorithm, selecting of filters from OpenCV library. Middlebury data set was selected as examples of images which were used for experiment. At the beginning filters were compared with using only on input images and then on all pyramid levels. For filters comparison average angular error and average endpoint error were used. As a result, one filtering method was selected for image filtering in Lucas-Kanade algorithm.