Error Characterization of Vision-Aided Navigation Systems
Abstract
The goal of this work is to characterize the errors committed by an Image Aided Navigation (IAN) algorithm that has been developed for use as a navigation tool in GPS denied areas. The filter under study was developed by the Air Force Institute of Technology's Advanced Navigation Technology center, and has been the focus of numerous research efforts. Unfortunately, these studies have all been based on single runs or simulations, and such results may not be indicative of the true filter performance. This problem extends to IAN publications in general; no analysis of IAN based upon a sizable real world data collection appears in the literature. This issue is addressed by applying Monte Carlo analysis methods to a 100 run data set collected using a joystick controlled robot outfitted with an inertial unit and stereo cameras. The averaged error magnitudes are found to be within 1 m RMSE. In addition, optimism in the filter computed covariance is verified. Finally, two instances of filter divergence are explored, with the causes being traced to feature matching errors. The results of this work will support future research efforts by providing a baseline measure of filter performance against which prospective enhancements may be compared.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2013
- Accession Number
- ADA584184
Entities
People
- Daniel A. Marietta
Organizations
- Air Force Institute of Technology