Integrating Building Information Modeling and Panoramic Structure-from-Motion for Accurate Camera Pose Estimation

In this study, we present a novel approach for combining Building Information Modeling (BIM) and panoramic photogrammetry-based Structure-from-Motion (SfM) to achieve accurate camera pose estimation in architectural scenes. The fusion of BIM and SfM information addresses the limitations of individual methods: the former offers global positioning, but it suffers from suboptimal accuracy; while the latter provides accurate relative positioning, it lacks scaling and global positioning. Our method consists of four key steps: (1) computationally efficient global positioning of panorama images in the BIM model using indoor semantic skymasks to generate probability distributions, (2) relative positioning estimation from the panoramic SfM process, (3) rough alignment of the SfM reconstruction with the BIM positioning using generalized Procrustes analysis (GPA), (4) refinement of the camera pose using non-linear least-squares optimization. We evaluate the performance of our proposed method using the real-world dataset of panoramic images capturing architectural scenes and compare the refined camera poses with ground truth. The results demonstrate camera positioning accuracy of fewer than 0.6 meters when compared to using BIM or panoramic SfM individually. This research highlights the potential benefits of fusing SfM and BIM modalities, paving the way for more accurate and efficient camera pose estimation pipelines in the architecture, engineering, and construction (AEC) domain.