Doctorat LMD
Permanent URI for this collectionhdl:123456789/5602
Browse
Item VIoT Solution: Computer Vision and IoT Integration in Monitoring and Inspection for Energy Conservation(Faculté des sciences et de la technologie, 2026-04-09) Seniguer AbderraoufThis thesis presents the design and development of a lightweight vision-based device that integrates Computer Vision (CV) and the Internet of Things (IoT) for real-time assessment of indoor illuminance distribution in buildings. The proposed system is intended for integration into daylight harvesting systems to enhance energy efficiency and improve occupants’ visual comfort. Unlike traditional approaches that rely on single photosensors, sensor grids, or High Dynamic Range (HDR) imaging, the proposed method uses visual information captured by a camera combined with machine learning techniques to estimate spatial illuminance levels across the observed scene. A dedicated data collection process was conducted using a controlled experimental setup, producing a dataset of approximately 12,000 samples covering illuminance levels from 0 to 2000 lux. From captured images, small pixel regions were extracted and associated with ground-truth measurements obtained from a calibrated lux meter. The illuminance mapping process is based on a two-stage machine learning pipeline composed of a Random Forest classifier and a Multilayer Perceptron regressor. This architecture enables accurate illuminance estimation while maintaining low computational complexity, allowing the entire system to operate directly on an embedded Raspberry Pi platform without cloud processing. In addition to illuminance estimation, the system incorporates a visual discomfort detection module for glare analysis. The device communicates with a web-based monitoring application through an IoT network, enabling real-time visualization and remote interaction. Experimental validation conducted in laboratory and real classroom environments demonstrated a Mean Absolute Error Percentage of approximately 8.7% when compared to reference illuminance sensors. The main contribution of this work lies in proposing a practical and deployable alternative to HDR imaging and photosensor grids for indoor illuminance mapping, combining computational efficiency, real-time performance, and IoT integration for smart building applications.