High Fidelity Physics Simulation-Based Convolutional Neural Network for Automotive Radar Target Classification Using Micro-Doppler

Abstract

Detection and classification of vulnerable road users (VRUs) such as pedestrians and cyclists is a key requirement for the realization of fully autonomous vehicles. Radar-based classification of VRUs can be achieved by exploiting differences in the micro-Doppler signatures associated with VRUs. Specifically, machine learning (ML) algorithms can be trained to classify VRUs using the spectral content of radar signals. The performance of these models depends on the quality and quantity of the data used during the training process. Currently, data collection is typically done through measurements or low fidelity physics, primitive-based simulations. The feasibility of carrying out measurements to collect training data is typically limited by the vast amounts of data required and practicality issues when using VRUs like animals. In this paper, we present a computationally efficient, high fidelity physics-based simulation workflow that can be used to obtain a large quantity of spectrograms from the micro-Doppler signatures of VRUs. The simulations are conducted on full-scale VRU models with a 77 GHz, frequency-modulated continuouswave (FMCW) radar sensor model. Here, we collect the spectrograms of 4 targets; car, pedestrian, cyclist and dog at different speeds and angles-of-arrival. This data is then used to train a 5-layer convolutional neural network (CNN) that achieves nearly 100% classification accuracy after 5 epochs. Studies are conducted to investigate the impact of training data size, velocity and observation time window size on the accuracy of the CNN. Results from this study demonstrate how an accuracy of 95% can be realized using spectrograms obtained over a 0.2 s time window.

An Overview of Millimeter-Wave Radar Modeling Methods for Autonomous Driving Simulation Applications

“…This method avoids farfield radiation calculations on the target and achieves near real-time computing speeds by reducing the number of rays and their bounces, but it results in lower computational accuracy. Ushemadzoro Chipengo and Juan D. Castro, among others, employed Ansys's SBR+ technique to develop several echo models [69][70][71][72][73][74][75], simplifying the computational steps for rays on object surfaces in ray tracing, incorporating pre-training, and leveraging GPU computations to conserve time. Despite enhancing accuracy and significantly reducing computation time compared to traditional ray tracing, it regrettably fails to achieve real-time simulation speeds.…”

High Fidelity Physics-Based Simulation of a 512-Channel 4D-Radar Sensor for Automotive Applications

“…To ensure energy conservation, each ray is associated with a ray tube [36]. The launched and weighted GO rays are used to ''paint'' and propagate PO currents on the CAD representing the scene [20], [36], [39]. The PO currents are then re-radiated with the resulting fields contributing to the scattered field.…”

“…Simulation has emerged as an alternative way of obtaining synthetic radar returns for developing signal processing methods and detection algorithms [20], [31], [32], [33], [34], [35], [36], [37], [38], [39]. A reason for this is that, unlike measurement, simulation is cheaper and less time consuming.…”

“…Therefore, the use of full wave methods such as finite element method (FEM) becomes highly inefficient, if not impossible. Ray-tracing, asymptotic techniques employing different implementations of the shooting and bouncing ray (SBR) method have been used to address this challenge [20], [31], [32], [33], [34], [35], [36], [37], [38], [39]. A facet-based, hybrid simulation was conducted in [31] to determine the impact of multipath propagation on the radar cross section of low-flying targets in a maritime scene.…”

Comparative Analysis of Audio Processing Techniques on Doppler Radar Signature of Human Walking Motion Using CNN Models

“…In the domain of speech recognition systems, Abdel-Hamid et al [29] pioneered the use of CNN-based models for phone recognition. Furthermore, studies such as [30,31] have indicated that CNNs have been successful in the task of human motion classification as well.…”