AI-based mobile mapping targets self-driving cars

Article By : Mitsubishi Electric

Mitsubishi Electric has developed an automated system that uses AI to create and renew precise three-dimensional maps.

Mitsubishi Electric has developed an automated mapping and extraction of transitions in mapping landscape system based on artificial intelligence (AI) and its own Mobile Mapping System (MMS), which generate and renew precise 3D maps that provide static information of roads and surrounding objects needed for autonomous driving.

Automated driving in Japan is expected to evolve from advanced driver assistance systems (ADAS) to automatic-driving level 3 (conditional autonomous operation) between 2019 and 2020, creating further demand for related systems. However, automatic driving systems will require combinations of in-vehicle sensors as well as dynamic maps, of which the biggest challenges would be to keep the map information constantly up to date.

Mitsubishi Electric said its automated mapping technology uses AI to quickly create precise, accurate three-dimensional maps. Only necessary information, such as road markings and traffic signs, is extracted from laser-point clouds and camera data measured and collected by MMS. Mitsubishi Electric's MMS provides 3D positional information of roads and roadside structures with an absolute precision within 10cm or less, which is collected via a system consisting of laser scanners, cameras and GPS antennas, while driving. AI improves the precision of extraction and recognition of the only data necessary, resulting in 10 times faster map creation compared to industry-standard manual creation. The system also costs less than conventional methods.

 

Figure 1: Mitsubishi's difference extraction technology is able to distinguish differences and changes where characteristic points do not match.
 

Mitsubishi Electric is using its difference extraction technology for earlier establishment of the dynamic map itself and more efficient updating and maintenance at a faster pace. By automatically extracting characteristic points of past data and the latest laser-point cloud data measured with MMS, the difference extraction technology is able to distinguish differences and changes where characteristic points do not match. With the technology, maintenance of dynamic maps and the updating of precise 3D maps can be accomplished much faster by automatic extraction of only the points that has changed, compared to updating the entire map each time.

Mitsubishi Electric plans to sell software utilising this automated mapping and difference extraction technologies to map publishers, including Dynamic Map Planning, in October. The software will be used for the creation of 3D maps of expressways in Japan.