A Novel Sparse Geometric 3-D LiDAR Odometry Approach

Abstract

Localization is a fundamental prerequisite, no matter whether a single robot or multirobot system, where light detection and ranging (LiDAR) odometry has attracted great interest with accurate depth information and robustness to illumination variations. In this article, a novel 3-D LiDAR odometry approach based on sparse geometric information is proposed. Different from geometric map-based 3-D LiDAR odometry methods with point features, we concern significant line and plane features based on eigenvalues of neighboring points. Furthermore, line-to-line and plane-to-plane associations instead of point-to-line and point-to-plane associations are adopted, and the problem of high computation complexity for scan-to-map matching module caused by point feature is solved. The proposed approach can not only guarantee the accuracy of pose estimation but also reduce computation complexity. Experiments on the public KITTI dataset and an outdoor scenario demonstrate the effectiveness of our approach in terms of accuracy and efficiency.