Abstract
High-precision and low-cost lane-level map are required from autonomous driving systems and advanced driving assistant systems. This paper proposes an automatic map synthesis method from laser point clouds that are scanned by a Mobile Mapping System (MMS). In this method, we apply a gradient image processing to detect map line segments. Results of the comparison of auto-extracted map and handmade map for the evaluation achieved over 80% extraction rate with an accuracy of 90%. We apply this system to a vehicle localization problem by matching camera images and the map and achieved sub-meter accuracy.
