SafeSight: Privacy-Preserving AI Passive Monitoring System for Situational and Health Awareness

Real-time monitoring systems are being increasingly utilized as an option to protect vulnerable populations, but current solutions heavily depend on sensors that reduce effectiveness and create privacy concerns.The proposed SafeSight system addresses this gap by applying an array of models to perform scene understanding, object detection, action recognition, and motion anomaly detection to enable contextual privacy masking.We evaluated SafeSight with benchmark datasets (UCF50, GMDCSA) and live video feeds.It achieves up to 0.867 F1 score for sedentary activity recognition and over 90% scene context accuracy within household environments.The use of artificial intelligence through deep learning and vision language models revolutionizes live video analysis to make accurate decisions with temporal scene analysis, static and dynamic events, and privacyprotecting contoured filters for sensitive locations.SafeSight could feasibly expedite alerts while monitoring high-risk situations where medical attention is required immediately, eventually being scaled to larger scenarios for healthcare, education, and public safety.