A Review on Fall Detection Techniques for Elderly People with Video Surveillance

Falls are a significant health issue worldwide, particularly for elderly people, contributing to a high number of injuries and deaths each year. The World Health Organization reports that over 684,000 fatalities annually are attributed to falls, with older adults being the most affected. This has led to an increasing need for reliable fall detection systems to minimize the adverse effects on both physical and mental well-being. This paper reviews and categorizes various fall detection techniques, including wearable device-based systems and emerging computer vision techniques. Wearable solutions like accelerometers and gyroscopes are cost-efficient but face challenges like user compliance and potential false alarms. Conversely, vision-based methods that use deep learning models, such as Convolutional Neural Networks (CNNs) and Spatio-Temporal Graph Convolutional Networks (ST-GCNs), offer improved detection capabilities. After surveying various fall detection systems for elderly people with video surveillance and alert notification, an in-depth comparison of advanced fall detection models is carried out, evaluating their performance on datasets like UR Fall Detection, NTU RGB-D, and Multi-Camera Fall datasets, where accuracy rates of up to 99% have been achieved.