Doppler-Spectrum Feature-Based Human-Vehicle Classification Scheme Using Machine Learning for an FMCW Radar Sensor.

Eugin Hyun,Youngseok Jin

doi:10.3390/s20072001

Abstract

In this paper, we propose a Doppler-spectrum feature-based human–vehicle classification scheme for an FMCW (frequency-modulated continuous wave) radar sensor. We introduce three novel features referred to as the scattering point count, scattering point difference, and magnitude difference rate features based on the characteristics of the Doppler spectrum in two successive frames. We also use an SVM (support vector machine) and BDT (binary decision tree) for training and validation of the three aforementioned features. We measured the signals using a 24-GHz FMCW radar front-end module and a real-time data acquisition module and extracted three features from a walking human and a moving vehicle in the field. We then repeatedly measured the classification decision rate of the proposed algorithm using the SVM and BDT, finding that the average performance exceeded 99% and 96% for the walking human and the moving vehicle, respectively.

Highlights

In the commercial market, radar sensors are applied to various platforms and applications, such as vehicles, robots, and drones, as well as medical, electronic, and safety applications [1]
We employ a radar front-end module consisting of a 24-GHz FMCW transceiver, used in previous
We proposed a human–vehicle classification scheme using a Doppler-spectrum feature-based SVM and BDT for a commercial FMCW radar system

Summary

Introduction

In the commercial market, radar sensors are applied to various platforms and applications, such as vehicles, robots, and drones, as well as medical, electronic, and safety applications [1]. In the field, because the scattering points of the Doppler spectrum returned from a walking human are highly variable with every measurement instance, the classification performance may be limited As another typical FMCW based method [14], different features of the range and velocity profiles were applied to classify humans and vehicles. In order to consider that the echo power of a walking human can vary due to magnitude fluctuations of the received signal, we propose a third feature, x3 The extraction of this feature is carried out during the “magnitude feature extraction” step in the right part of Figure 3.

Concept

Measurement Results

Measurement

We optimized the SVM and parameters

13. Examples proposed feature feature based based on on the the Doppler

Conclusions