Fall Detection System for Elderly People Using LSTM

Abstract

Falls are a leading cause of injury among older adults, requiring reliable real-time detection systems that can intervene promptly. This study proposes a detection system as a wearable device integrated with a six-axis Inertial Measurement Unit for sensing and a Raspberry Pi Pico microcontroller using a Long Short-Term Memory network for inference. Volunteers simulated elderly daily activities and fall cases to address practical constraints of real-world elderly data acquisition. The proposed double-layer LSTM architecture achieved an accuracy of 97.8%, compared to 92.19% obtained with the single-layer configuration, upon comparative evaluation, with corresponding reductions in false-positive and false-negative rates in confusion matrix analysis. The study highlights the potential of deep learning-based wearable technologies deployed on low-power hardware to improve elderly safety. The findings support the integration of wearable sensors and recurrent neural networks for monitoring and timely intervention for individuals with high chances and at high risk of falls. Future work could focus on the expansion of datasets, the integration of additional sensors, and testing in real-world environments.