Image to Lidar Registration Using Image Feature Matching

Abstract

Registration of the image respect to the point cloud scanned with lidar is a necessary procedure in various fields such as robotics, computer vision, virtual reality, and etc. Since the descriptors of the 2D image feature and the feature point of the 3D point cloud have different properties, it is difficult to directly match them. In this paper, instead of directly matching 2D feature points and 3D keypoints, we use the matching information between the cubemap image generated from LiDAR scanned point clouds and feature point of input images. We show the results of the proposed algorithm through experiments.