Creation of high definition map for autonomous driving within specific scene

Abstract

The high-definition map (HD Map) is one of the essential components for autonomous driving (AV) systems, which is also helpful to realize the intelligent transportation system. For autonomous vehicles, the HD Map is closely connected with the modules of high-precision localization, environment perception, and decision-making control in the AV systems. Although many mapping companies have launched HD Map products, such as HERE, Tomtom, Civil Maps, Baidu and NavInfo, there is currently no globally unified standard for HD Map. And these map companies are almost aimed at large areas such as regions and cities, and few of them are aimed at specific scenes. In addition, the mapping and maintenance costs of these HD Maps are also very expensive. In this paper, we develop a HD Map in a specific scene, which can not only better serve the AV application in the scene, but also reduce the cost of mapping, maintenance and updating. First, we give the definition classification of specific scenes. Next, we put forward a novel multi-client and multi-layer HD Map architecture for specific scenes. Finally, we introduce the key technologies for mapping of each layer, and verify the accuracy of the positioning layer by comparative experiment of relative accuracy.