For many years we have been dealing with the statistical analysis of geo-referenced data to support and improve the virtual testing and design of vehicles. Since the assistance and automation functions used in vehicles are becoming increasingly complex, classic testing and design procedures are increasingly reaching their limits.
Current approaches to describing the logic of a road network, for example, often fail to capture complex peripheral cases, as they are ubiquitous in reality. This includes, for example, incomplete road markings or defective asphalt. Real assistance systems must, however, achieve a safe driving condition even if there is no road marking. This must already be taken into account in the development process.
Real environmental data as a basis
The “VMC Road-and-Scene Generator” software package currently being developed at the ITWM enables the virtual development and testing of automation systems on the basis of real environmental data. The process works as follows: Using traditional statistical methods, we determine a representative city and then record it as a 3D point cloud using the institute’s own measurement vehicle REDAR. Using machine learning techniques, we then analyze and classify the measured data; relevant objects such as vehicles, lanes, road markings, buildings, etc. are automatically identified.
Automated data analysis and classification
This information provides the decisive contribution to an exact sensor simulation, as additional attributes such as material properties, reflection and absorption properties for different electromagnetic wavelengths etc. are now available for each object and for each measuring point. The data analysis and classification is largely automated, making the overall process highly efficient.