BoW3D: Bag of Words for Real-Time Loop Closing in 3D LiDAR SLAM (2208.07473v2)

Abstract: Loop closing is a fundamental part of simultaneous localization and mapping (SLAM) for autonomous mobile systems. In the field of visual SLAM, bag of words (BoW) has achieved great success in loop closure. The BoW features for loop searching can also be used in the subsequent 6-DoF loop correction. However, for 3D LiDAR SLAM, the state-of-the-art methods may fail to effectively recognize the loop in real time, and usually cannot correct the full 6-DoF loop pose. To address this limitation, we present a novel Bag of Words for real-time loop closing in 3D LiDAR SLAM, called BoW3D. Our method not only efficiently recognizes the revisited loop places, but also corrects the full 6-DoF loop pose in real time. BoW3D builds the bag of words based on the 3D LiDAR feature LinK3D, which is efficient, pose-invariant and can be used for accurate point-to-point matching. We furthermore embed our proposed method into 3D LiDAR odometry system to evaluate loop closing performance. We test our method on public dataset, and compare it against other state-of-the-art algorithms. BoW3D shows better performance in terms of F1 max and extended precision scores on most scenarios. It is noticeable that BoW3D takes an average of 48 ms to recognize and correct the loops on KITTI 00 (includes 4K+ 64-ray LiDAR scans), when executed on a notebook with an Intel Core i7 @2.2 GHz processor. We release the implementation of our method here: https://github.com/YungeCui/BoW3D.

• The paper introduces BoW3D, a novel method for real-time full 6-DoF loop closing in 3D LiDAR SLAM, addressing pose drift effectively.
• It utilizes LinK3D features and a dynamic bag-of-words approach with a hash table for efficient 3D feature retrieval and geometric verification.
• Empirical tests on the KITTI dataset show superior performance, making BoW3D valuable for autonomous driving and robotics applications.

Overview of BoW3D: Bag of Words for Real-Time Loop Closing in 3D LiDAR SLAM

The paper "BoW3D: Bag of Words for Real-Time Loop Closing in 3D LiDAR SLAM" by Cui et al. presents a novel approach to address the challenges in loop closing for 3D LiDAR SLAM systems. Loop closing is essential in SLAM to ensure global consistency by recognizing previously visited locations, which helps in correcting accumulated pose estimation drifts. While similar techniques have been successful in visual SLAM through bag-of-words (BoW) models, the authors propose an innovative application of this strategy to 3D LiDAR data.

Methodology

BoW3D leverages LinK3D, a 3D feature extraction method, to efficiently recognize and correct loop closures. The system is embedded in a 3D LiDAR odometry framework, specifically A-LOAM, to enhance loop closing performance. Key components of their approach include:

• Feature Extraction: LinK3D features are generated from LiDAR point clouds. These pose-invariant descriptors enable accurate feature correspondence, a crucial aspect of effective loop closure.
• Bag of Words for 3D Features: The authors develop a BoW system specifically tailored for 3D LiDAR features. Unlike traditional visual BoW models requiring pre-trained vocabularies, BoW3D dynamically generates its database in real-time through a hash table, allowing efficient storage and retrieval.
• Loop Detection and Correction: Utilizing a hash-based retrieval mechanism, BoW3D identifies loop candidates and performs geometric verification using RANSAC and SVD techniques to ascertain and correct the full 6-DoF loop pose.

Empirical Evaluation

BoW3D’s performance was evaluated against several leading methods in LiDAR place recognition, such as M2DP, PointNetVLAD, and Scan Context. Tested on the KITTI dataset, the method demonstrated superior results in terms of $F_1$ max scores and Extended Precision on most sequences. Importantly, BoW3D stands out with its capability to correct the full 6-DoF pose, unlike some competitors providing only partial pose corrections.

In real-world applications embedded within a 3D LiDAR SLAM system, BoW3D substantially decreased positional drifts, correcting the challenges associated with cumulative errors in odometry. Runtime analysis validated the method’s efficiency, with average processing times enabling real-time application.

Implications and Future Prospects

The development of BoW3D represents a significant advancement in 3D LiDAR SLAM, particularly in areas involving long-term autonomic navigation where revisitation of sites is frequent. Its ability to handle full 6-DoF corrections in real-time could be of high interest in autonomous driving, robotics navigation, and augmented reality. The method's efficient implementation without reliance on pre-trained models or extensive computational resources underscores its practical applicability.

Future work could explore extending BoW3D's capabilities to include advanced relocalization and mapping tasks, thus augmenting the robustness and accuracy of SLAM systems further. Integrating more complex semantic data or expanding the approach to accommodate diverse environments more effectively could also provide promising avenues for research. The authors' open-sourcing of BoW3D offers an accessible tool for further innovation and integration into various robotic systems and applications.