| Author: | Song, Yang |
| Title: | Deep learning-based adaptive UWB/INS integrated navigation system via factor graph optimization for UAV indoor localization |
| Degree: | M.Sc. |
| Year: | 2020 |
| Subject: | Drone aircraft -- Control systems Navigation (Aeronautics) Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Mechanical Engineering |
| Pages: | xii, 127 pages : color illustrations |
| Language: | English |
| Abstract: | Due to the widely use of the rotorcrafts in civil applications, the highly accurate positioning is paid more attentions. The ultra-wide band (UWB) receiver plays a crucial role in navigating the unmanned aerial vehicle (UAV) in indoor areas, because of its low cost and low power consumption. However, the positioning accuracy of UWB is drastically affected by the infamous multipath effect. Therefore, the ultra wide band (UWB)/inertial navigation system (INS) integrated indoor navigation is an effective approach to reduces the positioning errors. The Extended Kalman Filter (EKF) is widely used to integrate the UWB with INS measurements in indoor localization. However, the positioning results are unsatisfactory, on account of the weighting between UWB and INS is difficult to tune appropriate. In this study, we propose a tightly-coupled UWB/INS integration navigation based on factor graph optimization (FGO). For the loosely-coupled integration, we employ the linear and nonlinear least square methods to obtain the well-performed single point positioning. The Allan-variance analysis is used to study the characteristics of IMU sensors and estimate the process noise covariance of INS. Besides, to reduce the computational load of nonlinear optimization in factor graph and improve the positioning accuracy in indoor test site, we employ the state-of-art IMU preintegration technique as constraint factor in graph optimization. Furthermore, an adaptive factor graph optimization is required to develop and adjust the UWB measurement noise covariance matrix with different UWB measurement environments. Based on deep learning technology, convolutional neural network is employed to design signal classification and noise prediction network. The signal classification model is trained with the real operation data covered most scenes, further utilized to classify the UWB signal as line-of-sight (LOS) and non-line-of-sight (NLOS). Then, in order to obtain the quantitative analysis of measurement noise, the prediction network is constructed to predict the fixed value of UWB measurement noise after classifying. Finally, the location performance of the proposed FGO method is compared with an Extended Kalman Filter (EKF) in different test scenario, including LOS and NLOS scenes. The results show that the proposed tightly-coupled UWB/INS integration method can realize the better positioning performance than that of the conventional EKF in LOS and NLOS combined indoor environments, ensure the safety of UAV operation in complex indoor environments. Meanwhile, the improved adaptive factor graph optimization (AFGO) is tested on the LOS and NLOS mixed scenario. The results indicate that the proposed AFGO using signal classification and noise prediction network can achieve better performance in positioning accuracy of complex indoor scene. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 5239.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 7.3 MB | Adobe PDF | View/Open |
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/200/10802