CorrI2P: Deep Image-to-Point Cloud Registration via Dense Correspondence

Motivated by the intuition that critical step of localizing a 2D image in corresponding 3D point cloud is establishing 2D-3D correspondence between them, we propose first feature-based dense framework for addressing image-to-point registration problem, dubbed CorrI2P, which consists three modules, i.e., feature embedding, symmetric overlapping region detection, and pose estimation through established correspondence. Specifically, given pair cloud, transform them into high-dimensional space feed resulting features detector to determine where overlap each other. Then use regions establish before running EPnP within RANSAC estimate camera's pose. Experimental results on KITTI NuScenes datasets show our CorrI2P outperforms state-of-the-art methods significantly. We will make code publicly available.