Privacy-Aware Deep Learning and Cybersecurity Challenges in Automated Fall Detection for Smart Elderly Care

Abstract

Older people have a high rate of falls, with more than 70% of falls in hospital emergency departments being among the elderly, and falls cause a significant amount of morbidity and mortality in older people. At the same time, automated facial pain expression recognition is becoming an important non-invasive tool in assessing patient discomfort in geriatric care. This review provides an overview of the most recent technologies and approaches developed for automated fall detection and facial pain expression recognition, particularly focusing on deep learning methods. Threshold-based algorithms, classical machine learning methods like Support Vector Machines (SVM), Random Forests, and Artificial Neural Networks and deep learning architectures like Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and hybrid CNN-LSTM models are been reviewed. The following key sensor modalities are covered: wearable inertial measurement units (IMUs, accelerometers and gyroscopes), vision-based (RGB and depth cameras), and ambient sensors. Published results are provided for benchmark databases like UR Fall Detection and UNBC-McMaster Shoulder Pain Expression Archive. Persistent challenges include lack of real-world fall data, demographic bias, privacy issues with vision-based systems, and multi-modal sensor fusion. Future research directions include the integration of edge and fog computing, the use of transfer learning, explainable AI, and IoT-powered monitoring systems.