Abstract
Abstract. In this paper, we present our approach for robust long-term visual localization in large scale urban environments exploiting street level imagery. Our approach consists of a 2D-image based localization using image retrieval (NetVLAD) to select reference images. This is followed by a 3D-structure based localization with a robust image matcher (DenseSfM) for accurate pose estimation. This visual localization approach is evaluated by means of the ‘Sun’ subset of the RobotCar seasons dataset, which is part of the Visual Localization benchmark. As the results on the RobotCar benchmark dataset are nearly on par with the top ranked approaches, we focused our investigations on reproducibility and performance with own data. For this purpose, we created a dataset with street-level imagery. In order to have independent reference and query images, we used a road-based and a tram-based mapping campaign with a time difference of four years. The approximately 90% successfully oriented images of both datasets are a good indicator for the robustness of our approach. With about 50% success rate, every second image could be localized with a position accuracy better than 0.25 m and a rotation accuracy better than 2°.
Highlights
Modern vehicle-based and portable mobile mapping systems with multi-camera sensor systems combined with state-of-the-art georeferencing techniques enable a large-scale acquisition of accurate street level imagery
With 92% of the localized images in the Coarse accuracy class and 56% in the High class (< 0.25 m and < 2°) the results on our own data outperform the results made on the RobotCar Seasons dataset
The Medium class (< 0.5 m and < 5°) with 60% of oriented images shows a drop of 10% compared to the results on the RobotCar Seasons dataset
Summary
Modern vehicle-based and portable mobile mapping systems with multi-camera sensor systems combined with state-of-the-art georeferencing techniques enable a large-scale acquisition of accurate street level imagery. The resulting georeferenced collections of indoor or street level imagery covering large building complexes, entire cities or even states provide a powerful basis for urban infrastructure management. They bear a great potential for accurate visual localization and 6DOF pose estimation – even in areas with no or only poor GNSS coverage. Visual localization using existing 3D image spaces as a reference, promises to address the task of sensor positioning and the task of determining the sensor pose If both tasks can be solved reliably and accurately, can the new imagery be integrated into the existing database and used for measurement and asset management tasks. We discuss the results, which demonstrate the capability of our visual localization approach to reliably and accurately determine 6DOF image poses in urban spaces
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
