Environmental sensing with LED arrays : data analysis for a proof-of-concept rig

Abstract

This thesis explores the dual functionality of a matrix of light-emitting diodes (LEDs) serving as both a lighting infrastructure and a sensing mechanism for occupancy detection in indoor environments. Leveraging their bidirectional characteristics, it investigates a visible light sensing (VLS) approach to infer object presence. This method can enable privacy-preserving, cost-effective, and scalable applications in intelligent buildings. To evaluate this concept, a custom-built 5 × 5 LED array rig was used to capture reflected light patterns under controlled conditions for object detection. The data was modeled as grayscale image segmentation and analyzed using two supervised deep learning architectures. These models were trained to infer occupancy states based on temporal and spatial variations in reflected light intensity. Experimental results highlight the system’s effectiveness, with the U-Net model achieving Intersection over Union (IoU) scores reaching 96%. These findings validate the feasibility of using LEDs as integrated lighting and sensing units, offering a discreet, low-cost solution for intelligent spatial monitoring.