Roth and his co-authors Deqing Sun and Michael J. Black, who are currently researching at Google and the Max Planck Institute for Intelligent Systems, respectively, convinced with their 2010 paper “Secrets of Optical Flow Estimation and Their Principles”.
Optical flow is about estimating motion from video. For each pixel it is determined where it is moving to. The optical flow is used, for example, in video compression or the semantic analysis of moving scenes. The authors analyzed optical flow algorithms that were successful at that time to better understand which components are essential for the accuracy of motion estimation.
It turned out that many algorithms contain a filtering step. However, this results in the algorithm no longer corresponding to the mathematical objective function on which the algorithm is actually based. But exactly this filtering step is important for the accuracy. Roth, Sun and Black were able to specify a new, extended mathematical objective function that can explain the filtering step and extend it even further.
The resulting findings helped researchers and developers to significantly increase the accuracy of motion estimation algorithms and to design new, advanced algorithms.