Real-time data-driven estimation of radar cross-section of vehicles

Radar technology is one of key elements integrated in autonomous driving systems and advanced driver assistance systems (ADAS). Development of ADAS requires extensive validation of the involved radar systems, and real test driving for the validation can be costly. Although promising as a replacement for real test driving, virtual test driving to simulate radar systems in detail using physics-based electromagnetic simulation techniques is time consuming. This paper describes a data-driven approach to reduce the computation time for such simulations. As an initial attempt to introduce a machine learning model to simulate electromagnetic reflection property of vehicles based on their shapes, convolutional neural networks (CNN) whose input and output are the vehicle shape and its radar cross-section (RCS), respectively, are trained. The correlation coefficient between the estimated and ground truth RCSs can be as high as around 0.8 while the computation speed is significantly increased with a speed-up of approximately 50 times with respect to the ray-tracing method.