Stereo-IA: stereo visual intensity alignment and beyond under radiation variation

Abstract

Stereo vision is a hot research topic at present, but due to the radiation changes, there will be a large intensity difference between stereo pairs, which will lead to serious degradation of stereo vision based matching, pose estimation, image segmentation and other tasks. Previous methods are not robust to radiation changes or have a large amount of calculation. Accordingly, this paper proposes a new stereo intensity alignment and image enhancement method based on the latest SuperPoint features. It combines the triangle based bearings-only metric, scale-ANCC and belief propagation model and has strong robustness to radiation changes. The quantitative and qualitative comparison experiments on Middlebury datasets verify the effectiveness of the proposed method, and it has a better image restoration and matching effect under the radiation changes.