An Effective Method for Fall Detection Based on YOLO in IoT Applications

Abstract

Falls in the elderly are a serious and common problem, often leading to injury and even death. The consequences of such accidents go beyond physical harm, often including mental and psychological suffering for both those affected and their families. Furthermore, the economic burden of falls on health care systems is enormous, as it includes medical costs, rehabilitation costs, and potential long-term care needs. There have been many studies conducted in an effort to detect falls for warning systems based on vision-based approaches. However, these methods face challenges such as low accuracy rate and high computational cost that are not suitable for Internet of Things (IoT) applications. Therefore, in this study, we propose two effective methods with two scenarios for the task of fall detection in IoT applications. The first method applies YOLOv8 Pose to detect people. Then, calculate the height and width of the bounding box and calculate the threshold based on the difference between them. The second method uses the YOLOv8 and YOLOv9 model to train a fall detection model on the fall detection dataset. When a fall is detected, a warning message including a fall image and time is sent to relatives using the telegram application. Experimental results demonstrate that both proposed methods based on YOLO achieve high accuracy in detecting human falls. This research offers meaningful solutions in practice and integration into IoT systems to detect falls early.