Doung Uk Kim, Shinhoo Kang, Journal of Internet Computing and Services, Vol. 26, No. 5, pp. 17-27, Oct. 2025
10.7472/jksii.2025.26.5.17, Full Text:  HTML
Keywords: Elderly Fall Detection, CCTV Video Analysis, Object Detection and Tracking, YOLOv8

Abstract

With the global increase in the elderly population, falls among seniors have become a significant socio-economic issue. In South Korea, where the elderly population is rapidly growing, prompt rescue and response in the event of a fall are crucial. Previous studies have explored various approaches, such as wearable devices, IoT sensors, and expensive camera equipment. However, these methods have limitations including the need for additional device wear, high installation costs, and maintenance burdens. In this study, we propose a system that overcomes these limitations by utilizing existing CCTV footage to detect elderly falls in real-time without the need for extra sensors. The proposed system employs a YOLOv8-based deep learning model, using the YOLOv8m and YOLOv8x models to detect and track both human objects and safety furniture(e.g., chairs, sofas, beds). By analyzing the spatial relationships and movement patterns between these objects, the system is able to identify fall events. Moreover, when a fall is detected, the system is designed to send an SMS alert to enable a rapid emergency response. Experimental results under five scenarios show that the proposed system achieves high detection accuracy and real-time responsiveness in static environments. These results indicate that analyzing the spatial relationship between safety furniture and human objects can be effectively utilized for fall risk detection. In future research, we plan to fine-tune a pre-trained model using fall images, integrate it with a time-series model to further enhance fall detection performance, and validate its effectiveness through evaluations in real-world user environments.

Statistics
Cite this article