Temporally evaluated optical flow: study on accuracy

In order for visual feedback systems to have quick response, an image sequence captured at a much higher frame rate than the video rate has recently been employed. However when optical flow, which is often used for vision-based control, is computed by the conventional methods using such an image sequence, the result contains significant error because there is almost no variance between any two successive image frames. To deal with this problem, we propose temporally evaluated optical flow, which we compute by measuring the time required for each pixel to move a fixed distance using temporally consistent shifts of each pixel. Experimental results show that the proposed method improves accuracy and is superior to the conventional methods in terms of robustness.