Abstract

In the future, vehicles will gather more and more spatial information about their environment, using on-board sensors such as cameras and laser scanners. Using this data, e.g. for localization, requires highly accurate maps with a higher level of detail than provided by today's maps. Producing those maps can only be realized economically if the information is obtained fully automatically. It is our goal to investigate the creation of intermediate level maps containing geo-referenced landmarks, which are suitable for the specific purpose of localization. To evaluate this approach, we acquired a dense laser scan of a 22 km scene, using a mobile mapping system. From this scan, we automatically extracted pole-like structures, such as street and traffic lights, which form our pole database. To assess the accuracy, ground truth was obtained for a selected inner-city junction by a terrestrial survey. In order to evaluate the usefulness of this database for localization purposes, we obtained a second scan, using a robotic vehicle equipped with an automotive-grade laser scanner. We extracted poles from this scan as well and employed a local pole matching algorithm to improve the vehicle's position. Document type: Part of book or chapter of book

Abstract

High precision ground truth data is a very important factor for the development and evaluation of computer vision algorithms and especially for advanced driver assistance systems. Unfortunately, some types of data, like accurate optical flow and depth as well as pixel-wise semantic annotations are very difficult to obtain. Document type: Part of book or chapter of book