[Paper Review] A fast, complete, point cloud based loop closure for LiDAR odometry and mapping

Jiarong Lin, Fu Zhang|arXiv (Cornell University)|Sep 25, 2019

Robotics and Sensor-Based Localization25 references45 citations

TL;DR

The paper introduces a fast, rotation-invariant loop closure method for LOAM that uses 2D histograms of keyframe features to detect loops, followed by map alignment and pose graph optimization, all open-sourced.

ABSTRACT

This paper presents a loop closure method to correct the long-term drift in LiDAR odometry and mapping (LOAM). Our proposed method computes the 2D histogram of keyframes, a local map patch, and uses the normalized cross-correlation of the 2D histograms as the similarity metric between the current keyframe and those in the map. We show that this method is fast, invariant to rotation, and produces reliable and accurate loop detection. The proposed method is implemented with careful engineering and integrated into the LOAM algorithm, forming a complete and practical system ready to use. To benefit the community by serving a benchmark for loop closure, the entire system is made open source on Github

Motivation & Objective

• Motivate and address long-term drift in LiDAR odometry and mapping (LOAM).
• Develop a fast, complete loop closure pipeline that can operate with LiDAR data in real time.
• Provide a rotation-invariant, robust loop detection mechanism based on 2D histograms of keyframe features.
• Integrate loop closure into LOAM with map alignment and pose graph optimization.
• Open-source the system and datasets to serve as a benchmark for point cloud based loop closure.

Proposed method

• Represent each keyframe with two 2D histograms (plane and line features) derived from rotated, direction-normalized feature directions.
• Compute a rotation-invariant feature direction for each cell via eigen-decomposition of local plane/line covariance and derive a global rotation R from plane feature directions.
• Compare new keyframe histograms to the history database using normalized cross-correlation to detect potential loops with thresholds (0.90 for plane, 0.65 for line).
• Align the detected loop by registering the new keyframe map to the global map using edge-to-edge and plane-to-plane features, with a distance check (<0.1 m) for acceptance.
• Perform pose-graph optimization after loop closure using ceres-solver and update all cells in the map accordingly.
• Provide an open-source LOAM extension with datasets and code on GitHub.

Experimental results

Research questions

• RQ1 Can a 2D histogram-based representation of keyframes achieve fast and reliable loop closure for LiDAR-based SLAM?
• RQ2 Is rotation invariance achievable for point-cloud loop detection using plane/line feature directions?
• RQ3 How effective is map alignment and pose graph optimization in correcting drift after loop detection within LOAM?
• RQ4 What is the computational performance of the proposed loop closure pipeline on desktop and onboard platforms?
• RQ5 Does the proposed system generalize across large-scale outdoor and indoor LiDAR datasets?

Key findings

• The loop closure method detects loops by comparing 2D histograms of keyframes using normalized cross-correlation with thresholds (0.90 for planes, 0.65 for lines).
• The system achieves fast loop detection and maps alignment suitable for real-time operation on both desktop and onboard hardware (as detailed in timing results).
• Pose-graph optimization following loop closure significantly reduces long-term drift in LOAM maps.
• The method demonstrates accurate alignment of the point cloud map with external references (e.g., Google maps) after loop closure.
• The approach is implemented as an open-source LOAM extension with datasets and code available on GitHub.
• Rotation-invariance evaluation shows plane features maintain similarity under rotation; line features remain distinguishable across rotations.