A Fall Risk Detection Model for Infants While Sleeping based on Convolutional Neural Networks

Abstract

Falling from a bed is a common problem among infants, often leading to serious injuries such as head trauma, fractures, and even long-term neurological damage. According to data from the World Health Organization (WHO), falls are a leading cause of unintentional injuries among children, especially infants. To prevent these incidents, an effective early detection system is needed. Traditional approaches, such as motion sensors and surveillance cameras, have been employed to monitor infant movements and detect fall risks. However, sensor-based systems face limitations in terms of accuracy and coverage area. As an alternative, computer vision techniques have shown rapid advancements in recent years, with Convolutional Neural Networks (CNNs) proving to be highly effective in recognizing visual patterns, including human motion and posture detection. In this study, we propose a CNN-based model to detect the risk of infants falling from a bed while sleeping. The CNN architecture is designed to accurately recognize movements indicative of fall risks, such as approaching the edge of the bed or sudden changes in posture. Our contributions include (1) the design of a CNN architecture that supports effective and efficient training for fall risk detection, and (2) the creation of a dataset to classify infants as either safe or at risk of falling. Experimental results demonstrate that our proposed system achieves high accuracy in detecting potential fall risks, providing a reliable solution for infant safety monitoring.