SGLC: Semantic Graph-Guided Coarse-Fine-Refine Full Loop Closing for LiDAR SLAM

TL;DR

SGLC is introduced, a real-time semantic graph-guided full loop closing method, with robust loop closure detection and 6-DoF pose estimation capabilities, and is integrated into a SLAM system, eliminating accumulated errors and improving overall SLAM performance.

Abstract

Loop closing is a crucial component in SLAM that helps eliminate accumulated errors through two main steps: loop detection and loop pose correction. The first step determines whether loop closing should be performed, while the second estimates the 6-DoF pose to correct odometry drift. Current methods mostly focus on developing robust descriptors for loop closure detection, often neglecting loop pose estimation. A few methods that do include pose estimation either suffer from low accuracy or incur high computational costs. To tackle this problem, we introduce SGLC, a real-time semantic graph-guided full loop closing method, with robust loop closure detection and 6-DoF pose estimation capabilities. SGLC takes into account the distinct characteristics of foreground and background points. For foreground instances, it builds a semantic graph that not only abstracts point cloud representation for fast descriptor generation and matching but also guides the subsequent loop verification and initial pose estimation. Background points, meanwhile, are exploited to provide more geometric features for scan-wise descriptor construction and stable planar information for further pose refinement. Loop pose estimation employs a \mbox{coarse-fine-refine} registration scheme that considers the alignment of both instance points and background points, offering high efficiency and accuracy. Extensive experiments on multiple publicly available datasets demonstrate its superiority over state-of-the-art methods. Additionally, we integrate SGLC into a SLAM system, eliminating accumulated errors and improving overall SLAM performance. The implementation of SGLC will be released at https://github.com/nubot-nudt/SGLC.

Figures (4)

• Figure 1: Visualization of loop closing using our method. (a) Loop closure detection, it shows a reverse loop on the KITTI 08 sequence found by our approach even with significant changes in the position and orientation. (b) Semantic node correspondences for geometric verification and initial loop poses estimation. The blue lines indicate the node correspondences and red spheres represent the estimated instance center. (c) Final alignment for loop correction.
• Figure 2: The framework of SGLC. It first builds a semantic graph for foreground instances and then generates LiDAR scan descriptor considering both the topological properties of the semantic graph and the appearance characteristics of the background. The LiDAR scan descriptor is utilized to retrieve loop candidate scans from the database. Following this, geometric verification is performed on each loop candidate to filter out false loop closure, with the key step utilizing the instance node descriptors for robust sparse node matching. Finally, a coarse-fine-refine registration scheme is employed to estimate the precise 6-DoF pose.
• Figure 3: The qualitative comparison of loop pose estimation on the KITTI dataset using overlap-based loop pairs. Dashed ellipses are directly annotated on the registration results, while solid boxes indicate local magnification. The top-left corner displays the ground truth distance and yaw angle difference of the loop pair.
• Figure 4: The trajectory of A-LOAM odometry (left) compared to with integrating BoW3D loop closing method (middle) and integrating our approach (right) on the KITTI 00 sequence.