Hybrid Approach for Reflective Surfaces Reconstruction Using Automotive Radar

Automotive radar is the main sensor enabling autonomous driving and advanced driver-assistance system (ADAS) capabilities. Its major role is to provide high-accuracy information on the vehicle surrounding. However, in dense urban environments, multiple objects limit radars' line-of-sight and degrade their performance due to the multipath phenomenon, which generates ghost targets. Efficient radar's operation in the presence of occluding and reflecting surfaces requires their accurate reconstruction. This work proposes a hybrid approach combining the model with the deep neural network to reconstruct the reflected surfaces from the received radar echoes. Accurate reconstruction of reflecting and occluding surfaces is expected to enable automotive radar non-line-of-sight operation and multipath-induced ghost targets mitigation. The ability of the proposed approach to reconstruct the reflective surface is evaluated via simulations, and its superiority over the conventional estimation approaches is demonstrated.