| Author: | Huang, Feng |
| Title: | Dynamic object-aware LiDAR odometry in urban areas : from single to cooperative navigation |
| Advisors: | Hsu, Li-ta (AAE) Wen, Weisong (AAE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Department: | Department of Aeronautical and Aviation Engineering |
| Pages: | xxvi, 186 pages : color illustrations |
| Language: | English |
| Abstract: | Robust and precise positioning is critical for the autonomous system with navigation requirements. In recent years, Light detection and ranging (LiDAR) odometry have been extensively studied to achieve this goal. Satisfactory performance of LiDAR odometry (LO) can be achieved in sub-urban areas with abundant environmental features and limited moving objects. However, the performance is significantly degraded in challenging urban canyons with numerous moving objects. Moreover, the LO is subjected to drift over time. Global Navigation Satellite Systems (GNSS) can provide reliable absolute positioning in open areas and serve as a complement to LO. However, GNSS performance is often degraded in urban areas due to signal reflections caused by surrounding structures. In this thesis, we developed new methods to mitigate the impact of outliers in LiDAR odometry, enhancing positioning performance for autonomous driving in urban environments. First, we evaluated several popular and widely studied LO pipelines using datasets collected from urban canyons in Hong Kong, presenting the results in terms of both positioning accuracy and computational efficiency. We concluded three key factors that affect LO performance in urban canyons: ego-vehicle dynamics, moving objects, and the degree of urbanization. Second, we conducted an in-depth study on how to improve LO performance with the existence of large amounts of dynamic objects using deep learning-based techniques and point-wise discrepancy images. LO performance was further improved by applying object reweighting in highly dynamic scenarios. Third, we proposed a LiDAR-aided cycle slip detection method for GNSS-RTK, which effectively identifies cycle slips in carrier-phase measurements by leveraging consecutive relative pose estimates provided by LO. Furthermore, we present roadside infrastructure-assisted navigation in urban areas. First, we explore the use of roadside LiDAR to provide accurate states that serve as the global constraint in the LiDAR/Inertial odometry (LIO) graph-based optimization. Second, we present the use of consistent roadside double-differenced (DD) constraints provided by roadside GNSS are jointly optimized into the factor graph optimization. Third, we introduce an error-map-aided multi-sensor integrated system that utilizes error information collected by a sensor-rich autonomous vehicle. This error data are then uploaded to the roadside infrastructure, where it is subsequently distributed to other vehicles, which benefits the navigation performance of other vehicles. Numerous experiments are conducted using the onboard sensor platform and vehicle-infrastructure platform to validate the performance of the proposed method. The proposed dynamic object-aware LO significantly enhances positioning accuracy, achieving decimeter-level precision compared to the traditional meter-level accuracy in high-dynamic environments. With the assistance of roadside sensors, the proposed method achieved a 36.6% improvement in terms of absolute positioning accuracy compared to the state-of-the-art GNSS/LiDAR/INS integrated method. The evaluation results demonstrate that our proposed methods outperform the conventional positioning methods, providing accurate and reliable positioning and mapping for autonomous driving in urban canyons. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13737