Hybrid SVM-CNN Classification Technique for Moving Targets in Automotive FMCW Radar System

Abstract

Moving target classification is an essential ingredient to avoid accidents in autonomous driving systems. Recently, 77GHz automotive frequency modulated continuous wave (FMCW) radar has been popularly used to recognize moving targets due to its robustness to weather and light conditions, but the reliable classification of object types has been proved to be quite challenging. In this paper, a hybrid SVM-CNN method that jointly exploits support vector machine (SVM) and convolutional neural networks (CNN) techniques is proposed for target classification in the automotive radar system. The proposed method makes full use of the features of the targets as well as Range-Doppler images of echo signal, and two-stage scheme in the method is able to leverage class-imbalance problem in the real automotive context. Experiments have verified the effectiveness and correctness of the proposed method.