Evaluation of 3D point cloud for autonomous vehicle（3D point cloud evaluation index for detecting the double structure）

Abstract

Localization in autonomous vehicles is an important technology, and the use of 3D point clouds, which provide accurate information on the road surroundings, has been attracting attention to help improve localization. In recent years, many methods for constructing 3D point clouds have been proposed for use in autonomous vehicles. However, 3D point clouds can be misaligned due to errors in measurement high accurate sensors, and so on, which causes the failure of localization. Therefore, it is important to confirm the accuracy of the 3D point clouds and the feasibility of high-accuracy localization in advance. The accuracy of 3D point clouds is often confirmed via simulations using sensor data collected by vehicles if the localization is sufficiently accurate. However, the applications of 3D point clouds are expanding, and it would be preferable to avoid using sensor data for misalignment detection. Therefore, in this paper, we propose an indicator to detect the location of the misalignment of a 3D point cloud constructed by Mobile Mapping System, using only the 3D point cloud as an indicator of convergence and similarity of the ground objects via matching. The effectiveness of the proposed method was confirmed by the evaluation test, which showed that the proposed method can detect positional shifts in 3D point clouds even at locations containing similarities among geological features.

Keywords

• localization
• map evaluation
• 3d point clouds
• normal distributions transform
• autonomous vehicle