Development of a Real-Time Fall Detection Prototype with Transition Dynamics Using an IR Camera and IoT Alert System

This paper presents the design, implementation, and evaluation of a real-time fall detection system intended to enhance safety monitoring for elderly individuals, with particular emphasis on night-time scenarios. The system integrates an infrared (IR) camera with the YOLOv8 deep-learning detector to localise the human body, while a rule-based transition-dynamics module interprets bounding-box geometry to confirm fall events. An Internet-of-Things (IoT) alert mechanism, comprising a local active buzzer and a Telegram notification service, provides immediate on-site and remote response. The prototype, deployed on a Raspberry Pi 4, achieved 84.62% live-deployment accuracy with an average detection latency of 0.8–1.2 s, while the underlying detector attained a mean Average Precision (mAP@50) of 98.4%. The IR imaging pipeline preserves occupant privacy through silhouette-based sensing and maintains effectiveness across illumination conditions. A comprehensive analysis of aspect-ratio thresholds, dynamic transition velocity, error sources, and end-to-end response times is provided. Experimental results confirm the feasibility of this non-intrusive, low-cost approach for continuous elderly care monitoring in residential environments.