Using Normal Vectors for Stereo Correspondence Construction

Abstract

In scenes of an outdoor environment, depth recovery by matching a pair of stereo images is not very successful due to the strong effect of noise and changing illumination. In contrast, the surface normal is relatively robust against the influence of noise or changing illumination compared to other frequently used features. In this paper, we propose a two-step approach to solve the 3-D depth recovery problem. In the first step, we use the intensity feature to execute a rough comparison. We then use the surface normal vector, which is a much more discriminating feature, as the search basis for the second step. In addition, the 3-D invariant nature of a surface normal improves the accuracy of the stereo image matching results. The maps reconstructed in our experiments on images of outdoor scenes show that our approach is indeed more efficient and accurate than conventional methods.