Vision-based navigation with reality-based 3D maps

Abstract

This research is focused on developing vision-based navigation system for positioning and navigation in GPS degraded environments. The main research contributions are summarized as follows: a. A new concept of 3D map, which mainly consists of geo-referenced images, has been introduced. In this research, it provides the map-matching function for vision-based positioning. b. A method of vision-based positioning with use of photogrammetric methodologies has been proposed. It mainly obtains geometric information of the navigation environment from the 3D map through SIFT based image matching and uses photogrammetric space resection to solve the position in 6 degrees of freedom. The algorithms have been tested in an indoor environment. The accuracy has reached around 10 cm. c. A multi-level outlier detection scheme for the vision-based navigation system has been developed. It mainly combines RANSAC with data snooping. The former one deals with high percentage of mismatches, while data snooping removes outliers from different sources in the least squares adjustment for both 3D mapping and positioning solution. d. The deficiency of using RANSAC for outlier detection in image matching and homography estimation has been identified. In this research, a novel method which combines cross correlation with feature based image matching has been proposed. It is able to evaluate the RANSAC homography estimation and improve the image matching performance. The method has been successfully applied to the vision-based navigation solution to find corresponding view from the database and improve the final positioning accuracy. e. The positioning performance of the system has been evaluated through the analysis of mathematical model and experiments. The focus has been on various image matching conditions/methods and their impact on the system performance. The strength and weaknesses of the system have been revealed and investigated. f. The vision-based navigation system has been extended from indoor to outdoor with corresponding changes. Besides camera, it also takes advantage of multiple built-in sensors, including GPS receiver and a digital compass to assist visual methods in outdoor environments. Experiments demonstrate that such system can largely improve the position accuracy in areas where stand-alone GPS is affected and can be easily adopted on mobile device