Low-Cost FMCW Radar Human-Vehicle Classification Based on Transfer Learning

Abstract

Detection and classification of moving targets is an essential feature in many applications like road surveillance systems, autonomous cars, and smart gate systems. Multi-chirp sequence Frequency Modulated Continuous Wave (FMCW) radars with a 2D FFT processing can be used to produce a Range-Doppler images (R-D maps) containing the signature of the target. However, in low-cost FMCW radars, these images suffer from many problems like low-resolution and ambiguity. Such problems can make the image look unrealistic as well as hard to process and classify. In this paper, we propose a human-vehicle classification method based on Transfer Learning. The classification is done by processing the R-D maps generated by a low-cost short range 24 GHz FMCW radar with a convolutional Neural Network (CNN). The adopted CNN succeeded to reach a 96.5% accuracy in discriminating humans from vehicles.