A method of scene flow estimation with bilateral filter and adaptive TV(Total Variation) penalty function

摘要

This paper presents a method for estimating the three dimensional motion information of every point in a scene between two time steps (scene flow). In order to improve the accuracy of scene flow estimation, we propose a novel method of bilateral filter and adaptive TV penalty function based on variational framework from a calibrated stereo image sequence. First, we use bilateral filter preserve the flow filed motion edges and remove outliers. Second, we use the adaptive TV penalty function which can select TV-L¹ or TV-L² penalty function automatically in different image region. In order to prove our method is more accurate, we use the real datasets and artificial synthesis datasets with ground truth. We evaluate the accuracy according to the error between the ground truth and the estimated flow. Experimental results show that our method can obtain more accurate dense scene flow.