Neural Network Model Automates Lighting Detection in 2D Floor Plans

A research team has developed a neural network-based model to automatically detect lighting symbols, identify light types, and extract associated text from 2D floor plans. The model, built on Mask RCNN, was trained on annotated images converted to COCO format. It achieved bounding box mean average precision (bbox_mAP) of 0.7596 and segmentation mAP (segm_mAP) of 0.7111, with higher scores at stricter IoU thresholds (bbox_mAP 50: 0.9850, segm_mAP 75: 0.9219). This tool aims to streamline architectural and construction workflows by reducing design time and enabling efficient estimation of lighting power requirements per floor. The research is detailed in a paper available on arXiv (ID: 2608.14317), announced as a cross submission. The model's performance suggests practical applicability for industries relying on floor plan analysis.