Edge-Deployable LSTM-Based Fall Detection on ESP32 with Accelerometer-Gyroscope Fusion for Elderly Safety

Abstract

Due primarily to weakened muscle strength and age-related health problems, physiological deterioration with aging causes reduced mobility and increased susceptibility to falls. To address this urgent problem, we propose the development of an Internet of Things (IoT)- enabled wearable device designed to identify falls in older adults and promptly notify emergency contacts to lower the risk of severe injury. Our system collects movement data in real time using high-precision motion sensors, which include a gyroscope and a 3-axis accelerometer. To distinguish between normal everyday activities and actual fall incidents, the data is analyzed using advanced machine learning techniques, specifically Long Short-Term Memory (LSTM) neural networks. The device prioritizes user comfort and techno- logical stability, allowing for a smooth and inconvenient-free integration into the user’s daily routine. Even in limited network conditions, the responsiveness of the system is enhanced by a single mobile application. Our approach demonstrates significantly higher fall detection accuracy compared to conventional low-complexity models. Future developments will expand the system’s capabilities to include on-going health monitoring (such as heart rate and oxygen saturation), gesture-based interaction, and customized recovery assistance through the mobile app, such as post-fall exercise instructions. Additionally, to improve proactive care and emergency response, predictive analytics will be used to identify prolonged periods of inactivity and possible fall hazards.