Graph-based robust 3D point cloud map merging approach for large scale

Linqiu Gui

doi:10.1007/s10586-024-04797-6

2024-10-28

PRE

DOI：10.1007/s10586-024-04797-6

原文链接

原文求助

摘要

En 中文

Autonomous driving is a crucial area of research and a key focus for industrial advancement, with the advancement of high-level autonomous driving technology depending significantly on creating precise 3D point cloud maps (PCM\\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\backslash$$\end{document}PCMs) of the driving environment. The challenge arises in constructing a comprehensive map in a single attempt, particularly in scenarios with constrained environments or limited system hardware resources. Consequently, the need to construct the PCM in multiple local areas through either multi-robot collaborative SLAM or time-sharing single-robot SLAM becomes imperative. The fusion of these local maps into a globally consistent map is achieved through overlapping area matching and pose graph optimization. However, error matching can pose a significant obstacle to graph optimization, leading to a notable reduction in system performance. Therefore, enhancing the robustness of these processes in the presence of false-positive matches is crucial. This paper introduces a graph-based robust 3D PCM merging approach for large-scale applications. Our system utilizes a classical two-step matching method to find the matching pairs between sub-maps: coarse matching with global descriptors and fine matching through point cloud registration. We apply spatial consistency detection to analyze the matches and determine the variance of residuals through the error propagation of the Special Euclidean Group. Based on the above and the clustering of matching pairs, we proposed a local and global two-step outlier removal module to filter out error matches, thereby improving the robustness of the PCM merging algorithm. Experimental results using KITTI and self-collected data demonstrate the effectiveness of our approach.

Keyword:

Pose graph

3D point cloud map (PCM) merging

Outlier removal

SLAM