Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

R-VoxelMap: Accurate Voxel Mapping with Recursive Plane Fitting for Online LiDAR Odometry

Haobo Xi, Shiyong Zhang, Qianli Dong, Yunze Tong, Songyang Wu, Jing Yuan, Xuebo Zhang

TL;DR

R-VoxelMap addresses outlier sensitivity and plane merging issues in voxel-based maps for LiDAR odometry by introducing a geometry-driven recursive plane fitting framework that uses an outlier detect-and-reuse pipeline and a distribution-based plane validity check. The method constructs a voxel map with per-voxel probabilistic planes stored in a recursive octree, enabling early extraction of large planes and reducing erroneous cross-plane merging, all within an IEKF-based pose estimator. Extensive experiments across KITTI, M2DGR, M3DGR, NTU VIRAL, and AVIA demonstrate improved localization accuracy with comparable runtime and memory to prior voxel-map approaches, validating robustness in both structured and unstructured environments. This work enhances the practicality of voxel maps for real-time LiDAR odometry by delivering higher accuracy and stable performance in diverse settings, with code planned for open-source release.

Abstract

This paper proposes R-VoxelMap, a novel voxel mapping method that constructs accurate voxel maps using a geometry-driven recursive plane fitting strategy to enhance the localization accuracy of online LiDAR odometry. VoxelMap and its variants typically fit and check planes using all points in a voxel, which may lead to plane parameter deviation caused by outliers, over segmentation of large planes, and incorrect merging across different physical planes. To address these issues, R-VoxelMap utilizes a geometry-driven recursive construction strategy based on an outlier detect-and-reuse pipeline. Specifically, for each voxel, accurate planes are first fitted while separating outliers using random sample consensus (RANSAC). The remaining outliers are then propagated to deeper octree levels for recursive processing, ensuring a detailed representation of the environment. In addition, a point distribution-based validity check algorithm is devised to prevent erroneous plane merging. Extensive experiments on diverse open-source LiDAR(-inertial) simultaneous localization and mapping (SLAM) datasets validate that our method achieves higher accuracy than other state-of-the-art approaches, with comparable efficiency and memory usage. Code will be available on GitHub.

R-VoxelMap: Accurate Voxel Mapping with Recursive Plane Fitting for Online LiDAR Odometry

TL;DR

R-VoxelMap addresses outlier sensitivity and plane merging issues in voxel-based maps for LiDAR odometry by introducing a geometry-driven recursive plane fitting framework that uses an outlier detect-and-reuse pipeline and a distribution-based plane validity check. The method constructs a voxel map with per-voxel probabilistic planes stored in a recursive octree, enabling early extraction of large planes and reducing erroneous cross-plane merging, all within an IEKF-based pose estimator. Extensive experiments across KITTI, M2DGR, M3DGR, NTU VIRAL, and AVIA demonstrate improved localization accuracy with comparable runtime and memory to prior voxel-map approaches, validating robustness in both structured and unstructured environments. This work enhances the practicality of voxel maps for real-time LiDAR odometry by delivering higher accuracy and stable performance in diverse settings, with code planned for open-source release.

Abstract

This paper proposes R-VoxelMap, a novel voxel mapping method that constructs accurate voxel maps using a geometry-driven recursive plane fitting strategy to enhance the localization accuracy of online LiDAR odometry. VoxelMap and its variants typically fit and check planes using all points in a voxel, which may lead to plane parameter deviation caused by outliers, over segmentation of large planes, and incorrect merging across different physical planes. To address these issues, R-VoxelMap utilizes a geometry-driven recursive construction strategy based on an outlier detect-and-reuse pipeline. Specifically, for each voxel, accurate planes are first fitted while separating outliers using random sample consensus (RANSAC). The remaining outliers are then propagated to deeper octree levels for recursive processing, ensuring a detailed representation of the environment. In addition, a point distribution-based validity check algorithm is devised to prevent erroneous plane merging. Extensive experiments on diverse open-source LiDAR(-inertial) simultaneous localization and mapping (SLAM) datasets validate that our method achieves higher accuracy than other state-of-the-art approaches, with comparable efficiency and memory usage. Code will be available on GitHub.
Paper Structure (14 sections, 12 equations, 5 figures, 7 tables, 1 algorithm)

This paper contains 14 sections, 12 equations, 5 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. System Overview
  5. R-VoxelMap Construction
  6. Plane Validity Check
  7. Map Update and Octree Reconstruction
  8. Point-to-Plane Matching for State Estimation
  9. Experiments
  10. Accuracy on KITTI
  11. Accuracy on LIO datasets
  12. Runtime and Memory Usage
  13. Stability Analysis
  14. Conclusions

Figures (5)

  • Figure 1: VoxelMap plane features suffers from outlier sensitivity (c, yellow-marked planes are erroneously shifted due to surrounding outliers), over-segmentation (d, white-boxed planar features are overly segmented, resulting in large uncertainty), and incorrect plane merging (d, red-marked regions show erroneous plane merging, generating large planar structures in areas that are not physically planar), all of which reduce localization accuracy. Our method effectively addresses these limitations.
  • Figure 2: System overview of the LiDAR-based odometry using R-VoxelMap. The red-highlighted modules represent the core of R-VoxelMap, a geometry-driven recursive map construction algorithm based on an outlier detect-and-reuse pipeline. The blue modules are specifically designed to support the fundamental functions of R-VoxelMap.
  • Figure 3: Comparison between R-VoxelMap construction (lower part) and VoxelMap construction (upper part). VoxelMap's plane parameters are sensitive to outliers under relaxed thresholds (upper left), while strict thresholds cause over-segmentation (upper right). Our recursive method uses outlier detect-and-reuse pipeline and plane validity check to effectively reduce these issues (see Algorithm 1).
  • Figure 4: Illustration of the point distribution-based plane validity check. (a) Obtain the best candidate plane from RANSAC result. (b) Discretize the plane into 2D grids with a fixed resolution, project inlier points onto the plane and count the number of points in each grid. (c) Cluster occupied grids and compute the total number of points for each cluster. (d) Select the optimal cluster and refit the plane using its points.
  • Figure 5: (a) and (b) Comparison of mapping results between the proposed method and VoxelMap on the M3DGR dataset (corridor2 sequence). (c) and (d) End-to-end error comparison between the proposed method and VoxelMap on the AVIA dataset (avia1 sequence). The proposed method achieves better mapping quality and localization accuracy.