DMLO: Deep Matching LiDAR Odometry

TL;DR

DMLO introduces a learning-based LiDAR odometry framework that explicitly incorporates geometric constraints, significantly improving accuracy and robustness in real-world noisy data compared to existing methods.

Contribution

The paper proposes DMLO, a novel framework combining deep feature matching with geometric constraints for LiDAR odometry, outperforming prior learning-based approaches.

Findings

DMLO outperforms existing learning-based methods on KITTI and Argoverse datasets.

DMLO achieves accuracy comparable to state-of-the-art geometry-based methods.

The framework effectively decomposes pose estimation into matching and transformation steps.

Abstract

LiDAR odometry is a fundamental task for various areas such as robotics, autonomous driving. This problem is difficult since it requires the systems to be highly robust running in noisy real-world data. Existing methods are mostly local iterative methods. Feature-based global registration methods are not preferred since extracting accurate matching pairs in the nonuniform and sparse LiDAR data remains challenging. In this paper, we present Deep Matching LiDAR Odometry (DMLO), a novel learning-based framework which makes the feature matching method applicable to LiDAR odometry task. Unlike many recent learning-based methods, DMLO explicitly enforces geometry constraints in the framework. Specifically, DMLO decomposes the 6-DoF pose estimation into two parts, a learning-based matching network which provides accurate correspondences between two scans and rigid transformation estimation with a close-formed solution by Singular Value Decomposition (SVD). Comprehensive experimental results on real-world datasets KITTI and Argoverse demonstrate that our DMLO dramatically outperforms existing learning-based methods and comparable with the state-of-the-art geometry based approaches.