Omni Point: 3D LiDAR-Based Feature Extraction Method for Place Recognition and Point Registration
- 1 January 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 9 (8) , 5255-5271
Abstract
In autonomous driving, place recognition is an essential component for successful mission execution, where Simultaneous Localization and Mapping (SLAM) is widely used in the system to estimate the vehicle's position by avoiding error accumulation. Various research efforts based on deep learning or handcrafted methods are currently underway to improve the place recognition performance. However, deep learning-based approaches pose certain challenges compared to handcrafted methods such as being less interpretable and often requiring extensive training time. In this paper, we propose a handcrafted method-based efficient feature extraction method for place recognition and point registration that aims to overcome these challenges while achieving high performance. The proposed method involves extracting robust feature points, generating powerful descriptors, and achieving performance levels similar to deep learning methods, while ensuring algorithmic versatility. Validation results using the KITTI dataset are also included in the paper, demonstrating exceptional performance even in demanding scenarios involving rotated loops. The results show an average loop closure accuracy of 95% or better, and an average pose estimation accuracy of 0.1 meters for translation and 0.15 degrees for rotation.Keywords
Funding Information
- National Research Foundation of Korea
- Korea government (NRF-2020R1A2C2013091)
This publication has 50 references indexed in Scilit:
- A Robust 2D Lidar SLAM Method in Complex EnvironmentPhotonic Sensors, 2022
- OverlapNet: a siamese network for computing LiDAR scan similarity with applications to loop closing and localizationAutonomous Robots, 2021
- An Overview of LiDAR Requirements and Techniques for Autonomous Driving2021
- Weighted Scan Context: Global Descriptor with Sparse Height Feature for Loop Closure Detection2021
- Appearance-Based Loop Closure Detection with Scale-Restrictive Visual Features2019
- SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences2019
- PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition2018
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation2017
- NetVLAD: CNN Architecture for Weakly Supervised Place Recognition2016
- Towards a swarm of agile micro quadrotorsAutonomous Robots, 2013