Hybrid Physics-Based and Data-Driven Approach to Estimate the Radar Cross-Section of Vehicles

Radar technology is one of the key technologies used in automotive scene recognition for autonomous driving systems and advanced driver assistance systems (ADAS). ADAS development requires the exhaustive validation of participating radar systems, and validation through actual test driving can be expensive. Although promising as a replacement for real test driving, virtual test driving is time-consuming owing to the fact that it simulates radar systems in detail, employing physics-based electromagnetic simulation techniques. This paper describes a hybrid physics-based and data-driven approach to reduce the computation time required for such simulations. The radar cross-section (RCS) of a vehicle is chosen as a target result obtained through electromagnetic simulations. The data-driven model is implemented with a cascade of two convolutional neural networks (CNNs) which are trained using ground truth data calculated with a physics-based ray-tracing method. The ray-tracing method is employed for generating both the training data and a part of the input to one of the CNNs. The correlation coefficient between the estimated and ground truth RCSs can be approximately 0.8 while the computation time is lower than 120 ms.

[1]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[2]  M. Clemens,et al.  Discrete Electromagnetism With the Finite Integration Technique - Abstract , 2001 .

[3]  Cheng Yang,et al.  Data‐driven projection method in fluid simulation , 2016, Comput. Animat. Virtual Worlds.

[4]  Iasonas Kokkinos,et al.  DensePose: Dense Human Pose Estimation in the Wild , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[5]  Robert L. Cook,et al.  Distributed ray tracing , 1998 .

[6]  K. Yee.  Numerical solution of initial boundary value problems involving maxwell's equations in isotropic media , 1966 .

[7]  Hermann Winner,et al.  Deep stochastic radar models , 2017, 2017 IEEE Intelligent Vehicles Symposium (IV).

[8]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[9]  Andrew L. Maas.  Rectifier Nonlinearities Improve Neural Network Acoustic Models , 2013 .

[10]  Ming Ouhyoung,et al.  On Visual Similarity Based 3D Model Retrieval , 2003, Comput. Graph. Forum.

[11]  Hujun Bao,et al.  GPU-Based Shooting and Bouncing Ray Method for Fast RCS Prediction , 2010, IEEE Transactions on Antennas and Propagation.

[12]  Oliver Hennigh,et al.  Lat-Net: Compressing Lattice Boltzmann Flow Simulations using Deep Neural Networks , 2017, 1705.09036.

[13]  Ken Perlin,et al.  Accelerating Eulerian Fluid Simulation With Convolutional Networks , 2016, ICML.

[14]  Bent Dalgaard Larsen.  Real-time Global Illumination by Simulating Photon Mapping , 2004 .

[15]  Michiel van Ratingen,et al.  Implementation of Autonomous Emergency Braking (AEB), the Next Step in Euro NCAP's Safety Assessment , 2013 .

[16]  Yasuyuki Matsushita,et al.  RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews from Unsupervised Viewpoints , 2016, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[17]  Nils Thürey,et al.  Data-driven synthesis of smoke flows with CNN-based feature descriptors , 2017, ACM Trans. Graph..

[18]  Sergey Ioffe,et al.  Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , 2015, ICML.

[19]  K. Torrance,et al.  Theory for off-specular reflection from roughened surfaces , 1967 .

[20]  Wei Li,et al.  Convolutional Neural Networks for Steady Flow Approximation , 2016, KDD.