TL;DR
DZLoop is proposed, a novel LiDAR-based loop closure detection method that leverages multiple structural density images and Zernike moments that generates a robust global descriptor that effectively captures diverse structural characteristics of the environment.
Loop Closure Detection is a fundamental component of any SLAM system. By performing place recognition, a robot can correct accumulated drift errors arising from odometry uncertainties during the mapping process. Numerous techniques have been proposed in the literature to address this task under different sensor configurations, including RGB cameras and LiDAR. Among these, LiDAR-based SLAM has gained substantial attention due to its robustness in outdoor environments and its invariance to illumination changes. However, LiDAR sensors inherently provide less texture information compared to cameras, introducing additional challenges for loop closure detection. One of the most widely adopted approaches in LiDAR-based SLAM is Scan Context, recognized for its simplicity and effectiveness. This method has been successfully integrated into a broad range of applications and algorithms. Nevertheless, its simplicity can also lead to reduced robustness when no supplementary verification mechanisms are employed. In this work, we introduce Gaussian Scan Context, an enhancement to the original Scan Context that incorporates statistical analysis of the input point cloud. This approach accounts for the distribution of points within each context bin. Experimental results demonstrate that this enhancement improves both robustness and overall performance, as supported by various metrics.
Simultaneous Localization and Mapping (SLAM) is a core technology for indoor mobile LiDAR mapping systems. In GNSS-denied environments, localization accuracy is often degraded by cumulative drift. Loop closure detection remains challenging under large viewpoint variations and limited scan overlap. To address this challenge, this paper proposes a loop closure detection method based on Key Plane Structures (KPS). Unlike point-based methods that are sensitive to sensor scanning angles, the proposed method exploits the geometric stability of indoor plane structures to construct invariant global descriptors. This representation improves the robustness of loop closure detection across different viewpoints and provides more reliable loop closure constraints for pose graph optimization. Extensive experiments on real-world datasets validate the effectiveness of the method. In scenarios with large viewpoint variations, the method reduces the absolute trajectory translation error to 0.248 m, significantly outperforming Scan Context. In scenarios with sufficient scan overlap, it achieves centimeter-level positioning accuracy. Overall, the proposed method reduces trajectory drift by approximately 68%–97% compared to the pure odometry baseline, demonstrating stable performance in indoor LiDAR SLAM.
LiDAR-based place recognition is fundamental to loop closure and global localization in SLAM. Bird’s Eye View (BEV) projection has emerged as an effective paradigm for 3D point cloud processing, yet rotation robustness remains expensive: prior approaches either discretize SO(2) at <inline-formula><tex-math notation="LaTeX">$\mathcal {O}(N_{R})$</tex-math></inline-formula> inference cost or require high-parameter dual-branch cross-attention. Polar BEV representations convert heading-induced 2D rotations into 1D cyclic translations, but introduce a systematic <italic>radial distortion</italic> unaddressed by existing architectures: expanding annuli with decreasing LiDAR point density cause feature activations to collapse at large radii, while global attention mixes geometrically incompatible cross-radius representations. We propose two lightweight mechanisms to address these failure modes: <bold>Radius-Aware CoordConv</bold> concatenates a deterministic radial coordinate channel to recover feature energy in sparse outer rings; <bold>Ringwise Circular Attention (RCA)</bold> confines self-attention to 1D azimuthal rings at each fixed radius, preventing cross-radius feature mixing by construction and reducing attention complexity from <inline-formula><tex-math notation="LaTeX">$\mathcal {O}((H_\theta W_{r})^{2})$</tex-math></inline-formula> to <inline-formula><tex-math notation="LaTeX">$\mathcal {O}(W_{r} H_\theta ^{2})$</tex-math></inline-formula>. The resulting single-stream network, PolarBEVPlace, achieves a mean Recall@1 of 94.0% across 14 sequences on KITTI, NCLT, and UrbanNav-HK at 2.30 ms per scan with 1.06 M parameters—a 5.8× latency reduction and 10.7× MAC reduction over BEVPlace++, outperforming or matching it on 13 of 14 sequences.
LiDAR localization and mapping play an important role in mobile robotics and autonomous driving systems, but points from dynamic objects in urban environments can be incorrectly associated with the local map during scan-to-map registration, introducing erroneous constraints into pose estimation and thereby degrading localization accuracy. Existing occupancy-based dynamic object removal (DOR) methods typically face two challenges: insensitivity to overlapped dynamic regions and high sensitivity to occlusions. To address these issues, we propose a LiDAR-inertial localization framework with plug-and-play DOR. The DOR module removes dynamic points before registration, thereby reducing such erroneous constraints while preserving reliable static structures for more stable scan-to-map associations. It combines occupancy difference detection, connectivity-based region completion, and spatiotemporal overlap discrepancy to improve the completeness of dynamic region detection in overlapped regions and reduce the false removal of static points caused by occlusions. In addition, a multilayer consistency check based on Kullback–Leibler (KL) divergence is introduced during scan matching to reject surface constraints with inconsistent distributions, thereby improving data association reliability. The proposed method is evaluated on the KITTI, ECMD, UrbanNav, UrbanLoco, and a self-collected campus dataset. Compared with representative SOTA baselines, including VoxelMap, PV-LIO, TRLO, Dynamic-LIO, and BTSA, our method achieves the lowest average absolute trajectory error (ATE) among the compared methods, with 1.91 m on KITTI and 1.41 m on the selected ECMD comparison subset, demonstrating improved localization accuracy across environments with different scales and dynamic levels.
Loop closure detection plays a crucial role in LiDAR-based simultaneous localization and mapping (SLAM) by reducing accumulated drift and ensuring global map consistency. However, in substation inspection environments, repetitive equipment structures, complex ground conditions, and the limited discriminative capability of conventional point cloud descriptors often lead to degraded loop closure performance. To address these challenges, this paper proposes a ground-aware loop closure detection method for substation inspection. Ground segmentation is first applied to separate ground and non-ground points. Inspired by the polar-coordinate representation of Scan Context, a ground-aware descriptor is then constructed to jointly encode the geometric properties of ground regions and the height characteristics of surrounding structures. Furthermore, a loop closure matching strategy based on ring key retrieval, sector key alignment, and descriptor similarity evaluation is designed to achieve robust place recognition under viewpoint variations.
Mainstream LiDAR simultaneous localization and mapping (SLAM) systems often face challenges such as cumulative localization drift and loop closure mismatches when operating in large-scale complex outdoor environments, making it difficult to maintain global map consistency. This paper aims to propose a LiDAR SLAM system that integrates ground constraints with semantic information to comprehensively enhance robustness and localization accuracy in complex scenarios. First, to address pose estimation errors caused by complex terrain, a two-step ground segmentation strategy is proposed. This method obtains reliable ground parameters through coarse extraction and fine fitting, and introduces ground constraint factors into the backend pose graph optimization, effectively suppressing cumulative system drift. Second, addressing the issue of perceptual aliasing in traditional geometric loop closure detection within dynamic or structurally similar scenes, a coarse-to-fine two-stage loop closure detection method is proposed. The first stage uses Scan Context for rapid candidate frame retrieval, while the second stage integrates semantic topological features with geometric distributions for precise matching and verification, thereby eliminating false matches and calculating high-precision 6-DoF loop closure poses. Experimental results on a quadruped robot platform and the KITTI public data set demonstrate that the proposed method significantly reduces trajectory errors while maintaining real-time performance, showing superior performance in handling slopes, dynamic environments and large-scale loop closure scenarios. This work introduces a robust outdoor LiDAR SLAM framework that effectively combines ground constraints with a novel coarse-to-fine semantic loop closure detection method to resolve elevation drift and perceptual aliasing in complex environments.
