Enhancing Elderly Safety: A Study on Vision-Based Monitoring for Abnormal Behaviour Detection

Abstract

This study addresses the crucial need for innovative approaches in supporting the well-being of the rapidly aging global population by focusing on Abnormal Behaviour Detection (ABD) in Activities of Daily Living (ADL) for the elderly. Emphasizing the inadequacies of traditional monitoring methods, our research employs computer vision to accurately identify abnormal behaviours such as changes in activity patterns and falls. Central to this study is the evaluation of state-of-the-art models that takes into the account of temporal information, including Convolutional Long Short-Term Memory (ConvLSTM), Long-term Recurrent Convolutional Network (LRCN), and Convolutional Neural Network - Gated Recurrent Unit (CNN-GRU). The key contributions of this study include the generation of synthetic abnormal behaviour data using the limited publicly available dataset as well as the analysis of the best model, ConvLSTM by enhancing the transparency and understanding of its predictive mechanisms. This research not only advances the academic understanding of elderly care in vision-based approach but also offers practical insights for enhancing safety and well-being in various living environments for older adults.