mFall: A Scenarios-Based Research on Human Fall Detection Leveraging Millimeter-Wave Sensing

The issue of falls among the elderly is widely recognized globally, as falls often imply serious health risks. With the rapid development of technology, millimeter-wave (mmWave) sensing, due to its non-contact, continuous monitoring capabilities and higher safety and resolution, shows great potential in fall detection applications. However, current research still shows significant deficiencies in distinguishing between falls and non-falls in special scenarios and complex environments. Our research focuses on these deficiencies, collecting authentic human point cloud data. This includes basic daily activities as well as other common indoor movements that could be confused with falls. Then, we also gather point cloud data for irregular human activities introduced by complex scenarios. We use a sliding time window and convert the point cloud sequences into feature maps, which are then fed into the Deep Residual Network (ResNet). This process handles the spatio-temporal information contained in the point cloud data, ultimately yielding binary classification results. The system we develop shows efficiency in distinguishing between fall events and non-fall events, significantly improving the accuracy and reliability of fall detection, thereby offering a safe and accurate health monitoring solution for independently living seniors.