Lightweight visual monitoring system

Abstract

This project developed a real-time fall detection system designed to enhance the safety of elderly individuals in home environments. The system integrates ODROID boards running MotionEyeOS for continuous video surveillance, YOLOv8 for accurate fall detection, and a Telegram bot for immediate notifications to caregivers. MotionEyeOS enables efficient motion detection and event-driven activation of the YOLOv8 model, minimizing power consumption. The YOLOv8 model, trained on the v4_yaml dataset, utilizes a heuristic based on bounding box dimensions to identify falls. Experimental results demonstrated variability in accuracy across different environments, highlighting the impact of environmental factors on system performance. The system's design prioritizes energy efficiency and real-time responsiveness, providing a practical solution for continuous monitoring. Future work will focus on enhancing the heuristic and incorporating adaptive confidence thresholding, pose estimation, and environmental adaptation to improve accuracy and robustness. Longitudinal testing and integration with smart home systems are also recommended to further enhance the system's capabilities and user experience.