An Edge-Based Internet of Things (IoT) Architecture for Privacy-Preserving Fall Detection in Elderly Care

Abstract

Falls are a leading cause of injury and death among older adults. Most fall detectors rely on wearable sensors that send data to the cloud, which adds delay, exposes sensitive movement data and stops working when the internet is unavailable. This paper presents an edge-based IoT system that processes all sensor data locally. An MPU6050 motion sensor connected to a NodeMCU ESP8266 microcontroller streams data over Wi-Fi to a local Flask server running a hybrid CNN-LSTM model that separates falls from normal activities. On live data from several participants the system reached 97.4% accuracy. End-to-end alert latency was 1.2 s, compared with 4.3 s for the same model deployed in the cloud. Raw motion data never leaves the local network, notifications are encrypted before they are sent, and the system keeps working during internet outages. Average current draw is about 120 mA, which is comparable to commercial wearables. The results show that edge processing can match cloud accuracy while giving faster response and stronger privacy, making it a practical option for fall detection in elderly care