A LiDAR Odometry With Multi-Metric Feature Association and Contribution Constraint Selection

Abstract

LiDAR-based simultaneous localization and mapping (SLAM) is crucial for achieving accurate pose estimation and map generation, thus serving as a foundational technology in the advancement of autonomous driving systems. In this letter, we introduce an accurate and robust feature-based LiDAR odometry method. Initially, we propose a feature extraction method centered on local extreme points, which capitalizes on the structural characteristics of local regions in LiDAR scans. Secondly, we purpose a multi-metric feature association approach for keyframe registration. This method leverages sparse and abstract geometric primitives to improve the accuracy and speed of keyframe matching. Additionally, Considering the varying impact of different metric features on pose constraints, an constraint contribution selection method is introduced to identify the most valuable features within the multi-metric feature set. Finally, the performance and efficiency of the proposed method are evaluated on the public KITTI, M2DGR, and The Newer College dataset, as well as our collected campus dataset. Experimental results demonstrate that the proposed method exhibits comparable performance compared to state-of-the-art LiDAR odometry methods across various scenarios.