Non-contact sensor-based human fall detection method

Abstract

With the exacerbation of global population aging, the issue of falls among the elderly is increasingly drawing attention from various sectors of society. Consequently, the development of an effective human fall detection system is crucial for timely identification of fall incidents and provision of emergency assistance. This paper proposes a fall detection method based on non-contact sensors, aiming to achieve real-time monitoring and accurate identification of fall events through advanced sensor technology and data processing algorithms. Initially, this study employs low-frequency ultra-wideband radar (UWB) as the primary tool for data acquisition, followed by the utilization of deep learning techniques to extract key dynamic features from radar data, thus providing a reliable foundation for fall event identification. Furthermore, this paper designs and implements a fall recognition model based on Convolutional Neural Networks (CNN), which, through training on a large amount of radar image data, learns to distinguish patterns between normal activities and falling behavior. The results demonstrate that the proposed method can effectively detect falls under various common daily activities, with a recognition rate as high as 99.338%.