Indoor Human Activity Classification Based on FMCW Radar Micro-Doppler Signatures and Deep-Learning Networks

Abstract

Radar-based human activity classification using micro-Doppler signatures is one of the active research problems gaining a lot of attention over the years. This is due to its application in many fields, such as security, surveillance, health monitoring, etc. Most of the earlier studies utilized cameras and other wearable sensors to capture human activity data. However, in health-related applications such as remote health monitoring of elderly persons, user's privacy may not be guaranteed. In this regard, radar sensors can offer the essential advantage of capturing human activity without exposing the user's body efficiently. Therefore, this paper developed an indoor human activity classification system using Frequency Modulated Continuous Wave (FMCW) radar micro-Doppler signatures and Convolutional Neural Networks (CNN). Test accuracy of 91.35% was obtained using our costumed-designed CNN model. But after performing transfer learning, we obtained a test accuracy of 94.84% and 95.81% using Alex-Net and VGG-Net models, respectively.