Weighted root mean square approach to select the optimal smoothness parameter of the variational optical flow algorithms

Variational methods are the most widely used approaches for optical flow computation. Many complicated algorithms have been proposed to improve their performance, yet little work has focused on how to select the optimal smoothness parameter λ of the variational optical flow algorithm itself. We present a weighted root mean square error method to automatically select the optimal smoothness parameter λ. Furthermore, we detail a scientific method for selecting the reference λ0 based on the quality of the frame, and propose an efficient brute-force approach to assign a group of λ, that will reduce the number of λ candidates to be tested by cutting down the search range. Experimental results validate the effectiveness of our methods.