Our team’s paper, “Building Component Segmentation-oriented Indoor Localization using BIM-based Synthetic Data Generation and Deep Learning,” was presented at the 42nd International Symposium on Automation and Robotics in Construction (ISARC 2025), held in Montreal, Canada, from 28 to 31 July 2025.
The study introduces an image-based approach to indoor localization that combines synthetic images generated from building information models (BIM) with deep learning. The approach identifies key building components, including floors, ceilings, walls, doors, and windows, and compares their spatial arrangement in real photographs with BIM-generated views to estimate where an image was taken.
By focusing on building components that remain relatively stable over time, the method reduces reliance on highly detailed models of furniture and decorations. A case study in an institutional building demonstrated promising localization performance even with a simplified BIM model. The research offers a foundation for supporting indoor navigation, facility management, and emergency response. Read the paper.
The paper was co-authored by Haomiao Zhao, Yifeng Sun, Chi Chiu Lam, Junxiang Zhu, Mi Pan, and Mun On Wong.