Radar-Based Fall Detection Using Deep Machine Learning: System Configuration and Performance

Automatic fall-detection systems, saving time for the arrival of medical assistance, have the potential to reduce the risk of adverse health consequences. Fall-detection technologies are under continuous improvements in terms of both acceptability and performance. Ultra-wideband radar sensing is an interesting technology able to provide rich information in a privacy-preserving way, and thus well acceptable by end-users. In this study, a radar sensor compound of two ultra-wideband monostatic units in two different configurations (i.e., vertical and horizontal baseline) has been investigated in order to provide sensor data from which robust features can be automatically extracted by using deep learning. The achieved results show the potential of the suggested sensor data representation and the superiority of the double-unit vertical-baseline configuration. Indeed, while the horizontal configuration allows to discriminate the body's position around the radar system, the vertical one discriminates the body's height that is more important for fall detection.