Intelligent Monitoring for Elderly Well-Being: Deep Learning-Based Activity Recognition for Fall Detection
Areeb Adnan Khan, Syed Muhammad Mustafa, Haania Siddiqui, Syed Talal Wasim, Syed Nouman Hasany, Muhammad Farhan · 2023
The aging population poses unique challenges in terms of healthcare and well-being, requiring innovative solutions to ensure the safety and quality of life for elderly individuals. This research focuses on applying deep learning techniques for activity recognition in tracking and monitoring the daily activities of elderly people. The current literature on the matter suggests that state-of-the-art models find it difficult to accurately distinguish between falling and laying down. The study proposes a unique solution by dividing the dataset into local and global tags with local tags representing labels that are based on the information of that particular frame while global tags represent labels of a frame that are the same for the entire video. By using this dataset structure, we employ two CNNs: EfficientNet, ResNet, and a Vision Transformer (ViT-B16). The method proposed achieved promising results, with EfficientNet successfully distinguishing between falling and non-falling events with high accuracy. The paper also discusses the remaining two models, their shortcomings, and potential solutions for future work. In conclusion, this research contributes to the field of elderly care by showcasing the potential of deep learning in real-time activity recognition, where intelligent monitoring systems adapt to the specific needs of elderly individuals, promoting their well-being and independence while providing essential support for caregivers and healthcare providers.