Title: Comparisons of SR-UKF Family for a Visual-IMU Tightly-coupled System Based on Tri-focal Tensor Geometry
Author(s): Maosong Wang, Wenqi Wu, Naser El-Sheimy
Published in: Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017)
September 25 - 29, 2017
Oregon Convention Center
Portland, Oregon
Pages: 3088 - 3101
Cite this article: Wang, Maosong, Wu, Wenqi, El-Sheimy, Naser, "Comparisons of SR-UKF Family for a Visual-IMU Tightly-coupled System Based on Tri-focal Tensor Geometry," Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017), Portland, Oregon, September 2017, pp. 3088-3101.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: Unscented Kalman filter (UKF) is one of the best estimation techniques for achieving approximation to nonlinear systems since it is easier to approximate a probability distribution through unscented transformation (UT). However, its computation cost is highly reliant on the number of UT sigma points. UKF has better characteristics than extended Kalman filter (EKF). For example, it is more accurate and derivative-free, which means that the filter does not require the computation of Jacobian matrices. rices. In this research, the performances of several kinds of UT methods (spherical simplex sigma set, minimum sigma set and simplex sigma set), which use less than 2n sigma points, are compared. Some of the more ambiguous formulas and conceptual issues connected to these UT methods are revised through theoretical analysis. Four distinctive ego-motion cases from KITTI open source data are used to evaluate the performance of these UT methods. The system used is a tri-focal tensor geometry based binocular stereo Vision/IMU tightly-coupled system. By using these UT methods, the Kalman filter flows are a kind of square-root, since square-root unscented Kalman filters (SRUKF) can guarantee the stability of the system. The structure of the Visual-IMU tightly-coupled system is in the form of error state, and the time updates of the state and the state covariance are directly estimated without using UT transformation; thus computational time is greatly reduced. Results show that the SRUKF methods based binocular stereo Vision/IMU mechanization has overall higher accuracy than the UKF based Mono/IMU mechanization. Both the spherical simplex sigma set square-root unscented kalman filter (SSSRUKF) and the minimum sigma set square-root unscented kalman filter (M-SRUKF) based binocular stereo Vision/IMU mechanizations are as accurate as the standard SRUKF based binocular stereo Vision/IMU method. Additionally, by using reduced sigma point SRUKFs, the computation load is greatly reduced, since the main computation cost of SRUKF lies in the number of sigma points.