A Convolutional Neural Network Model for Privacy-Sensitive Ultra-Wideband Radar-Based Human Static Posture Classification and Fall Detection

Abstract

A reliable fall detection system can enhance the safety of senior citizens by detecting falls in private areas, such as restrooms, where accidents may go unnoticed. This study aimed to create a static human posture recognition system with a possibility of extension for detecting falls in private areas. The system used ultra-wideband (UWB) sensors to detect human body gestures and analyze an individual's posture to determine a laydown posture, which is abnormal in restroom usage. UWB is capable of protecting human privacy because its output contains limited information. This study implemented a convolutional neural network (CNN) model that classified signals from an ultra-wideband sensor in a bathroom into four categories: standing, sitting, lying down, and nobody. This paper proposes a CNN classifier with an overall accuracy of 93%. These results demonstrate the capability of the proposed system to recognize static human posture in private locations.