Automated Generation of Road Marking Maps from Street-level Panoramic Images

Accurate maps of road markings are useful for many applications, such as road maintenance, improving navigation, and prediction of upcoming road situations within autonomously driving vehicles. This paper introduces a generic and learning-based system for the recognition of road markings from street-level panoramic images. This system starts with an Inverse Perspective Mapping, followed by segmentation to retrieve road marking candidates. The contours of all found segments are classified, after which a Markov Random Field is applied to adjust the resulting probabilities based on the surrounding context. Finally, the spatial placement of the found individual markings (e.g. shark teeth) is analyzed to retrieve the traffic situations (e.g. priority situations). This system is evaluated for priority, block, striped lines and pedestrian crossing markings, and is able to recognize 80-95% of the individual markings, and about 90% of the occurring situations (e.g. pedestrian crossings).