| Author: | Bai, Xiwei |
| Title: | Outlier-aware GNSS/INS/visual integrated navigation system in urban canyons |
| Advisors: | Hsu, Li-ta (AAE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Global Positioning System Artificial satellites in navigation Inertial navigation systems Navigation -- Technological innovations Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Aeronautical and Aviation Engineering |
| Pages: | xvi, 137 pages : color illustrations |
| Language: | English |
| Abstract: | Accurate and globally referenced positioning is essential for a wide of autonomous applications, such as unmanned aerial vehicles (UAV), augmented reality (AR), and autonomous ground vehicles (AGVs). The visual-inertial navigation system (VINS) attracts lots of attention in recent years, due to its cost-effectiveness and accuracy. In general, the VINS can provide satisfactory odometry estimation in an ideal scenario with stable illumination conditions and sufficient texture information. Unfortunately, the performance of the VINS is significantly challenged in complex urban scenarios due to the unexpected dynamic objects (e.g. moving vehicles, pedestrians). Moreover, only relative positioning can be provided by the VINS. Global navigation satellite systems (GNSS) can provide reliable absolute positioning in open areas but are challenged in urban canyons due to the signal reflections. This thesis starts with perception-aided VINS where deep neural network (DNN) segmentation is utilized to identify dynamic objects. Instead of directly removing the detected dynamic features arising from the moving objects, this thesis proposed a re-weighting scheme to de-weight the dynamic features but maintain its geometry contribution. Improved positioning performance is obtained compared with the directly dynamic feature exclusion which we believe is an interesting finding. However, the method relied on the accuracy of the DNN for object detection. To this end, we seek to find a more general way to mitigate the impacts of the outlier feature measurements. As an extension, we developed a visual feature uncertainty model by considering the quality of feature tracking and distribution of features. The results showed that the proposed uncertainty model can effectively mitigate the impacts of visual outliers. However, the method relies on the redundancy of the healthy visual feature measurements which can be challenged significantly in dense urban canyons. Can we detect the outlier features from the source, for example, detect the visual outlier features from the feature tracking process? To this end, we developed a graduated non-convexity (GNC) aided optical flow to reliably detect the visual outlier features from the source during the feature tracking, right before the feature correspondences are fed to the backend optimization. The performance of the VINS is significantly improved with the help of the proposed method using the challenging dataset collected in dense urban canyons of Hong Kong which shows the feasibility of the proposed method. Although the performance of VINS was improved with the proposed methods mentioned above, it was still subjected to inevitable drift over time. Different from the local VINS positioning technique, the GNSS can provide globally referenced and free-drift positioning solutions, with an accuracy of a few meters in light urban canyons. However, the GNSS signals suffer from huge challenges that make them unreliable, such as multipath, non-line-of-sight (NLOS), and satellite coverage fluctuation, in urban canyons. Inspired by the strong time-correlation within the raw GNSS carrier-phase measurements (for example, the carrier-phase shares the same ambiguity across multiple epochs), this thesis proposed a time-correlated window carrier-phase (WCP) model to exploit the high accuracy carrier-phase measurement. In particular, the left null space matrix is employed to eliminate the unknown ambiguity variables shared across multiple epochs. The results showed that the proposed WCP model can effectively smooth the trajectory for GNSS standalone positioning using factor graph optimization (FGO). With the help of the WCP model, the resistance against the GNSS outliers is also improved in some ways using the FGO. Inspired by the high complementariness between the GNSS and the VINS positioning, the integration of the VINS and GNSS measurements is investigated in this study. How would the visual outlier mitigation methods listed above and the WCP model contribute to the complementary integration of the GNSS and VINS? To answer this question, this thesis proposed a tightly-coupled integration of the GNSS/VINS where the pseudorange, Doppler, and carrier phase measurements are directly integrated with visual features and inertial measurement unit (IMU) measurements based on the factor graph optimization (FGO) framework. With the help of the global positioning from raw GNSS measurements and local constraints from the visual and IMU, the trajectory of the local visual-inertial navigation system can be smoothed, and the drift can be significantly alleviated. With the help of the visual outlier mitigation technique, the accuracy is further improved. According to the evaluated dataset collected from a middle urban area of Hong Kong, an accuracy of 1.9 meters can be obtained using the proposed integration scheme. A positioning accuracy of about 4 meters can be obtained even in a super-urbanized area in Hong Kong with dense buildings and dynamic objects. |
| 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/200/12492