Nav-by-Search: Exploiting Geo-referenced Image Databases for Absolute Position Updates

摘要

Vision-aided navigation has been demonstrated to be aviable GPS-denied navigation option. However, onedisadvantage of vision-aided navigation is that it drifts,making it difficult to complete some missions after longGPS outages. Adding geo-referenced position updates byrecognizing places in a database has the potential toprovide a drift-free, GPS quality navigation solution inGPS-denied environments. This is a challenging task dueto changes in scale, viewpoint and lighting between thequery and database images. The proposed Nav-by-Searchmethod, first uses the place recognition algorithm,recognizing previously visited or mapped places, thenintegrating its results with our navigation system. Ourappearance based place recognition work has focused onimproving accuracy, approaching real-timeimplementation. Our place recognition results are shownusing real-world, street-level imagery in a dynamicenvironment with occlusions and significant change inappearance and viewpoint. Integrated Nav-by-Searchresults are also shown demonstrating drift-free navigation.