Speaker Bio & Abstract
BiographySeyed Majid Azimi is an artificial intelligence researcher and electronics engineer. During his Ph.D. at the Technical University of Munich (TUM), he achieved several proofs of concepts including the direct multi-class lane-marking extraction from space as an essential component of HD maps. He has been working in the German Aerospace Center (DLR) since 2016 on developing AI algorithms in order to realize High-Definition maps from aerial and satellite data. He is currently the CEO of Ternow AI GmbH as the DLR spin-off to provide HD maps from space.AbstractRealization of High-Definition Mapping for AD Using Aerial and Satellite Data (AeroSat HD Mapping)Nowadays, High-Definition (HD) maps are an essential component several applications such as urban management, transportation and infrastructure monitoring, city planning. HD maps are currently being generated using mobile mapping vehicles, driving through each driving area to capture the locations of traffic or infrastructure related objects. Besides the advantages of this approach, there are several drawbacks including demanding cost and time as well as low GNSS reception particularly in urban areas, and last, but not least the very low update rate. Until very recently, as a problem, there were no alternatives or no automatic approaches to validate the HD maps created by this approach.
In this talk, I will present our solution for HD mapping using aerial and satellite (AeroSat) data. AeroSat HD map solution combines multiple novel AI algorithms to extract very detailed semantic information down to the lane markings and traffic lamp poles. In order to achieve absolute cm-level geolocation precision, we have developed a novel sensor fusion technology to fuse RADAR and optical satellite data. AeroSat HD map does not have GNSS limits and has a significantly higher update rate. This method can also be used to validate the mobile mapping data or be fused for the sake of completeness. Because satellites cover the entire earth, large-scale HD maps of countries can be created without putting a foot on their ground.
Using our technology, HD mapping using AeroSat data is now a new possibility as an alternative or complementary approach to mobile mapping. This approach can also be used to validate HD maps being created by the current approaches reducing the cost and time required to create HD maps over a large area.
Satellite imagery has a high update rate. Satellite images can be taken from a relatively large area worldwide at any time as soon as a map update is required.
> Using the DLR solution, we can extract traffic-related objects directly from aerial and satellite data without requiring no 3rd party information as a full-package solution.
The combination of optical and radar satellite data offers the possibility of cm-level object localization even in downtown areas without GPS reception limits.