Robustified, Multi-epoch Stochastic Constraints for Outlier Detection and Removal in Online Multi-sensor Inertial Navigation Systems

M.J. Veth, A. Soloviev, C. Yang, C. Taylor

Peer Reviewed

Abstract:	Recent advances in sensor and signal processing technologies have fueled the development of a number of novel techniques for extracting navigation and targeting information using signals not normally considered for navigation. Some examples of these signals include, but are not limited to, magnetic fields, radio/cellular signals, video and image sequences, sonar, step/gait detection, and laser ranging measurements. Many techniques for integrating new types of observations have been demonstrated in the literature and have shown the ability to provide high-levels of accuracy in environments where Global Navigation Satellite System (GNSS) signals are unavailable. This accuracy potential has resulted in a great deal of interest for both military and civilian applications. In this paper, a statistically-rigorous approach is presented to detect and remove outliers by leveraging observations over multiple epochs. The presence of non-Gaussian errors is implicitly assumed in the model and used to develop a robustified-Gauss-Newton statistical estimator. The robust cost function-based approach has the advantage of improved convergence over traditional gating techniques due to the convex nature of the statistical cost function. The estimation results are used to calculate a robustified Mahalanobis distance which is used to indicate statistical outliers. These outliers can then be removed or de-weighted prior to incorporation into a navigation algorithm. While the technique is applicable for any feature-based navigation algorithm, an image-aided consumer-grade inertial navigation example is used to illustrate the performance. The algorithm was evaluated using two simulated scenarios: one representing indoor navigation in a hallway environment and one representing a flight scenario. For the indoor scenario, the algorithm demonstrated a probability of outlier detection of 99.78% with a probability of false alarm of 0.3%. Similar results were observed for the flight scenario.
Published in:	Proceedings of the 27th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2014) September 8 - 12, 2014 Tampa Convention Center Tampa, Florida
Pages:	2212 - 2219
Cite this article:	Veth, M.J., Soloviev, A., Yang, C., Taylor, C., "Robustified, Multi-epoch Stochastic Constraints for Outlier Detection and Removal in Online Multi-sensor Inertial Navigation Systems," Proceedings of the 27th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2014), Tampa, Florida, September 2014, pp. 2212-2219.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In