An AI–IoT-Integrated Vision-Based Localized Energy Management System for Smart Classrooms

Conventional classroom energy management systems often treat the entire room as a single occupancy zone, leaving fans and lights on even when only a few seating areas are occupied. This paper integrates a vision-based, localized energy management framework that uses the Ultralytics YOLOv8n model for real-time, zone-based occupancy detection in a 3×2 grid. The captured frames are resized to 640 × 480 pixels and filtered using a polygonal Region of Interest to analyze the classroom seating area. People are detected within localized bounding boxes using a bottom-centre point; perspective transformation accounts for the camera's inclination and mounting angle; and each person is assigned to one of the six zones arranged in a logical 3-column, 2-row grid. The multithreaded software architecture includes a dedicated camera acquisition thread that protects the frame buffer, and inference runs on every third frame to maintain real-time responsiveness. Based on the zone-wise occupancy, states are stabilized by state persistence, and ON/OFF decisions are made independently for the fan loads in each zone. The system was evaluated and implemented in a single classroom scenario. The results demonstrate reliable person detection, perspective-corrected zone assignment, and consistent localization of load-control behavior; however, the findings are limited to the evaluated configuration and architecture and are intended to support future extensions toward multi-classroom deployment.