Real-Time Human Fall Detection and Alert System Using Autonomous Embedded Neural Network

Falls are a major safety risk for older adults and individuals with reduced mobility, making prompt detection essential to reduce the likelihood of serious outcomes. This paper presents a real-time fall detection system built around two ESP32-WROOM-32 microcontroller units (MCUs) arranged in a sender–receiver configuration. An MPU6050 inertial measurement unit (IMU) is connected to the sender via the I2C protocol to obtain motion data, which is subsequently transmitted using the ESP-NOW protocol. The receiver processes this data to perform activity inference using a pre- deployed Multilayer Perceptron (MLP) model trained and tested in Edge Impulse. Detection of a fall triggers the automatic dispatch of an SMTP email notification to caregivers. A testing accuracy of 82.53% demonstrates the system’s viability for autonomous, cloud-independent, and resource-efficient wearable health monitoring.