Thalagala, Ravindu Gayanga (2024) Towards efficient and consistent visual inertial navigation using matrix lie groups. Doctoral (PhD) thesis, Memorial University of Newfoundland.
Full text not available from this repository.
Abstract
This thesis proposes a novel, computationally efficient, and consistent two key-frame visual-inertial navigation system (2KF-VINS) to cope with the demands of practical VINS application across full-scale aerial platforms and drones. For validation of VINS, this thesis produces a unique aerial multi-sensor field dataset captured on a full-scale aerial platform and a drone. The thesis validates the proposed 2KF-VINS by assessing its performance using the collected field dataset and incorporates the modules necessary to address the challenges encountered during field dataset experiments. The 2KF-VINS algorithm presented is designed using a combined Lie group SE2(3) extended Kalman filter (EKF) design framework. The conventional 2KF-VINS filter is unobservable for translations along all three axes and rotation about the gravity direction. As a result, the filter suffers from estimation inconsistencies related to unobservable transformations of the estimation problem. The proposed combined Lie group SE2(3) framework remedies this issue by implicitly preserving the observability consistency property of the filter. Monte Carlo numerical simulations are used to validate the theoretical performance of the right − SE2(3) 2KF-VINS, along with experimental validation using the EuRoC miro aerial vehicle (MAV) dataset. Additionally, the proposed algorithm is compared with state-of-the-art multistate constrained Kalman filter (MSCKF), right invariant (RI-MSCKF), left − SO(3), and right − SO(3) 2KF-VINS versions with identical and realistic tuning parameters to validate the performance related to the accuracy, consistency, and computational speed of the method. The field dataset presented in this thesis is captured using a multi-sensor payload capable of evaluating visual-inertial (VI) and visual-inertial-LiDAR (VIL) algorithms. The dataset features flight distances ranging from 300m to 5km, collected using a DJI M600 hexacopter drone and the National Research Council (NRC) Bell 412 Advanced Systems Research Aircraft (ASRA). The dataset consists of hardware synchronized monocular images, IMU measurements, 3D LiDAR point clouds, and high-precision real-time kinematic (RTK)-GNSS based ground truth. Ten datasets were collected as ROS bags over 100 mins of outdoor environment footage ranging from urban areas, highways, hillsides, prairies, and waterfronts. The datasets collected facilitate the development of VINS, visual-inertial-LiDAR odometry and mapping algorithms, object detection, segmentation, and landing zone detection algorithms based on real-world drone and full-scale helicopter data. All the datasets contain raw sensor measurements, hardware timestamps, and spatio-temporally aligned ground truth. The intrinsic and extrinsic calibrations of the sensors are also provided, along with raw calibration datasets. A performance summary of state-ofthe- art methods applied to the datasets is also provided. The captured data, along with the required resources to use them, can be downloaded from the following link: https://mun-frl-vil-dataset.readthedocs.io/en/latest/. The 2KF-VINS algorithm validations presented include the filter performance measures in estimation accuracy and computation efficiency. Computational efficiency is measured based on the filter’s execution time, while estimation accuracy is evaluated by incorporating root mean squared error (RMSE) for position and orientation. Additionally, the percentage drift of position and orientation with respect to the distance travelled was also presented as an estimation accuracy measure. Two datasets, namely the DJI M600 lighthouse and Bell412-1 datasets, were selected for this evaluation to represent small-scale and full-scale platforms, respectively. Various challenges faced by the 2KF-VINS in this field experiment validation were identified, including position drift during hovering conditions, sub-optimal filter re-initialization, and inconsistencies in the smoothing approach. Auxiliary modules in the VINS pipeline were introduced to mitigate these issues, which include hovering condition detection, re-initialization using back-end smoothing algorithms, and improving covariance estimates of the overall pipeline using filtering methods.
| Item Type: | Thesis (Doctoral (PhD)) |
|---|---|
| URI: | http://research.library.mun.ca/id/eprint/16646 |
| Item ID: | 16646 |
| Additional Information: | Includes bibliographical references (pages 135-151) --Restricted until July 30, 2025 |
| Keywords: | aerial, navigation, autonomous, helicopter, drone |
| Department(s): | Engineering and Applied Science, Faculty of |
| Date: | April 2024 |
| Date Type: | Submission |
| Library of Congress Subject Heading: | Inertial navigation systems; Aerial rockets; Drone aircraft; Algorithms--Design and construction |
Actions (login required)
 |
View Item |
