Classification of High Resolution Automotive Radar Imagery for Autonomous Driving Based on Deep Neural Networks

Abstract

Recently, there has been an increase in research for deep neural networks that perform object classification for self-driving vehicles using electro-optical sensors. Public optical datasets and classification algorithms that enable such development already exist, however, only radars can provide robust sensing in adverse conditions, when the optical systems may fail. The development of high resolution radar is necessary in order to approach radar imagery classification with neural networks in a similar way with optical images. Therefore, this paper presents a method of classification of six different roadside targets in low-THz imaging radar, using Convolutional Neural Networks. The present results confirmed that neural networks can also be successfully employed for low-THz radar imagery classification with high resolution and this method can be applicable for implementing an all-weather sensing system for autonomous vehicles.