Reliable road information based on high-resolution map data is essential for automated driving. However, the material still has a great headroom for improvements, especially off the highways. In order to close these gaps and create HD maps covering the entire area, the TRI-AD experts use various methods.
In cooperation with other partners, TRI-AD considered building layout information for the development of the high-resolution maps. This information is derived from the cameras of ordinary vehicles and from satellite images without the use of survey vehicles. The Toyota researchers also took into account vehicle data from the open “Automated Mapping Platform”. With the information collected, the maps can also be updated more quickly. In addition, the map material can be extended to other areas, and the costs for installation and maintenance are significantly reduced.
Together with Maxar Technologies, a partner experienced in space infrastructure, and the Japanese IT service provider NTT DATA, TRI-AD has, for example, produced HD maps using high-resolution commercial satellite images. The necessary information was automatically extracted, and cars, shadows and occlusions of buildings were analyzed and removed. With an accuracy of up to 25 cm, the maps proved useful for automated driving in a sample of 23 points in Tokyo and six cities outside Japan.
The Toyota researchers also carried out camera-based HD mapping with US map company Carmera. This involved using in-car dashcams to capture key road features such as road markings, traffic signs and traffic lights. A relative accuracy of 40 cm was achieved for the most important navigation features. For features that were only obtained from camera recognition, this represents a significant advance. Not only can the information be recognized, validated and fed into updates within minutes. The use of globally distributed vehicle cameras also enables scaling for next-generation mapping.
Navigation specialist TomTom and European map specialist HERE also participated in the feasibility study. The collected vehicle data could be automatically integrated into the map material of the two providers – in some cases even in real time.