Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MS-Mapping: Multi-session LiDAR Mapping with Wasserstein-based Keyframe Selection

Xiangcheng Hu, Jin Wu, Jianhao Jiao, Wei Zhang, Ping Tan

TL;DR

This paper presents MS-Mapping, a novel multi-session LiDAR mapping system that incorporates an incremental mapping scheme to enable efficient map assembly in large-scale environments and introduces a real-time keyframe selection method based on the Wasserstein distance.

Abstract

Large-scale multi-session LiDAR mapping is crucial for various applications but still faces significant challenges in data redundancy, memory consumption, and efficiency. This paper presents MS-Mapping, a novel multi-session LiDAR mapping system that incorporates an incremental mapping scheme to enable efficient map assembly in large-scale environments. To address the data redundancy and improve graph optimization efficiency caused by the vast amount of point cloud data, we introduce a real-time keyframe selection method based on the Wasserstein distance. Our approach formulates the LiDAR point cloud keyframe selection problem using a similarity method based on Gaussian mixture models (GMM) and addresses the real-time challenge by employing an incremental voxel update method. To facilitate further research and development in the community, we make our code\footnote{https://github.com/JokerJohn/MS-Mapping} and datasets publicly available.

MS-Mapping: Multi-session LiDAR Mapping with Wasserstein-based Keyframe Selection

TL;DR

This paper presents MS-Mapping, a novel multi-session LiDAR mapping system that incorporates an incremental mapping scheme to enable efficient map assembly in large-scale environments and introduces a real-time keyframe selection method based on the Wasserstein distance.

Abstract

Large-scale multi-session LiDAR mapping is crucial for various applications but still faces significant challenges in data redundancy, memory consumption, and efficiency. This paper presents MS-Mapping, a novel multi-session LiDAR mapping system that incorporates an incremental mapping scheme to enable efficient map assembly in large-scale environments. To address the data redundancy and improve graph optimization efficiency caused by the vast amount of point cloud data, we introduce a real-time keyframe selection method based on the Wasserstein distance. Our approach formulates the LiDAR point cloud keyframe selection problem using a similarity method based on Gaussian mixture models (GMM) and addresses the real-time challenge by employing an incremental voxel update method. To facilitate further research and development in the community, we make our code\footnote{https://github.com/JokerJohn/MS-Mapping} and datasets publicly available.
Paper Structure (9 sections, 8 equations, 2 figures, 1 algorithm)

This paper contains 9 sections, 8 equations, 2 figures, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. Motivation and Challenges
  3. Problem Formulation
  4. Wasserstein-based Keyframe Selection
  5. Keyframe Selection Formulation
  6. Wasserstein Distance for Map Dissimilarity
  7. Voxel Gaussian and Incremental Update
  8. Real-time Keyframe Selection
  9. CONCLUSIONS

Figures (2)

  • Figure 1: Measuring differences between two point cloud maps using Wasserstein distance. The Wasserstein distance includes residuals of both the mean and variance, capturing both global and local differences between distributions. After voxelizing the point cloud maps, the number of points in each voxel represents its "mass," allowing the calculation of Wasserstein distance between voxel pairs. The overall GMM map difference is the average Wasserstein distance of all voxels. Thus, by comparing the average Wasserstein distance before and after adding a point cloud frame to the map, it can be determined if the frame is a keyframe.
  • Figure 2: Incremental update and maintenance of the GMM map. After initializing the GMM map, each new point cloud frame is transformed into the map coordinate. The Gaussian parameters of the voxels in the GMM are then updated point by point according to their indices. Voxels outside the radius range are removed. This process reflects the influence of the new frame on the overall map distribution.