Enhancing Elderly Safety Through AI-Based Human Activity Recognition for Affordable Community Healthcare

Abstract

The rapid growth of the ageing population has increased the demand for affordable and effective monitoring solutions that enable older adults to live independently while ensuring their safety. Falls remain a major cause of injury and hospitalization among elderly individuals, highlighting the need for timely detection and intervention. Existing monitoring approaches, such as wearable devices and conventional surveillance systems, may be limited by user compliance, implementation costs, and privacy concerns. This study proposes an artificial intelligence (AI)-based non-contact human activity recognition system for affordable elderly monitoring using a low-cost vision-based approach. The system employs a standard webcam, OpenCV, and MediaPipe-based pose estimation to extract human skeletal landmarks, while an interpretable rule-based classification algorithm is used to recognize four activities: standing, walking, sitting, and falling. Experimental evaluation was conducted using 80 activity samples collected under controlled indoor conditions. The system achieved an overall recognition accuracy of 80%, with class-specific accuracies of 100%, 80%, 75%, and 65% for standing, sitting, walking, and falling, respectively. The findings demonstrate the feasibility of combining AI-based pose estimation with lightweight rule-based classification for real-time activity monitoring using affordable hardware. Local video processing also reduces the need to transmit visual data to external servers, thereby supporting greater data privacy. Although further refinement is required to improve fall detection, the proposed approach provides a practical foundation for accessible elderly monitoring and has the potential to support caregivers, independent ageing, and community-oriented healthcare.