Hypercomplex Neural Network Based Elderly Activity Recognition For Intelligent Healthcare Systems

Elderly care is becoming increasingly critical due to the rising global aging population. Automatic monitoring systems that can intelligently recognize human activities in real-time offer promising solutions for enhancing elderly safety, independence, and well-being. This research proposes a novel approach for elderly activity recognition using Hypercomplex Neural Networks (HCNNs) integrated with the PoseNet model for real- time video analysis. The system architecture includes user registration, login, and video upload. The uploaded video is processed to extract frames, and human pose landmarks are detected using the PoseNet model. These pose features are then fed into a Hypercomplex Neural Network, which leverages quaternion-valued representations to enhance spatial-temporal learning for better activity classification. The system classifies elderly activities into five categories: walking, sitting, running, fighting, and sleeping. Unlike traditional CNNs, HCNNs can model multidimensional features more compactly and with higher efficiency, improving classification accuracy in complex movements. This research aims to support assisted living environments by enabling real-time monitoring of elderly individuals, detecting abnormal behaviors (e.g., fighting or falling asleep unexpectedly), and issuing alerts when necessary. Our experimental evaluation demonstrates that HCNN outperforms conventional models in terms of recognition accuracy, robustness, and response time. The system thus presents a promising step toward intelligent and autonomouselderly care solutions.