Edge AI-Based Fall Detection with Standard RGB Cameras

Falls among elderly individuals represent a significant public health concern due to their frequency and severe consequences, such as injuries, reduced independence and increased healthcare demands. Current fall detection methods primarily rely on wearable sensors, which face usability and acceptance challenges, particularly among cognitively impaired users. This paper proposes a fully non-invasive fall detection solution based on vision technology and edge computing, using standard RGB cameras and low-power hardware. The system leverages YOLOv8, a lightweight and efficient convolutional neural network, to perform real-time on-device inference without requiring user cooperation. Experimental evaluations conducted in real-world residential care environments demonstrated high accuracy and reliability with rapid response times. The proposed approach provides a practical, cost-effective and user-friendly alternative to traditional wearable-based systems, suitable for a variety of healthcare and residential settings