Toward Automated SLAM-to-BIM for Digital Twin Applications: BIM-Assisted 360-Degree Camera-Based Visual SLAM Using Reference Point–Based Bundle Adjustment

Abstract The construction industry faces poor productivity issues due to frequent design changes and other disruptions during construction progress. To improve productivity, many researchers have been exploring the adoption of digital twins in industry. For digital twins, accurately aligning scanned as-built data with as-designed data is essential. To further advance this movement, this study proposes a building information modeling (BIM)-assisted 360-degree visual simultaneous localization and mapping (SLAM) framework that simultaneously performs autonomous alignment between 360-degree visual SLAM and BIM, and reference point–based refinement of camera pose estimation. To achieve this, BIM-registered reference points are directly incorporated into the bundle adjustment of a 360-degree visual SLAM pipeline, allowing camera poses to be estimated and optimized directly in the BIM coordinate system. Unlike post-hoc alignment methods, the proposed method integrated reference point information within SLAM optimization, allowing three-dimensional (3D) camera pose estimation and autonomous BIM alignment to be achieved simultaneously using only 360-degree video data and a limited number of reference points. Experimental results demonstrated that the proposed method achieves a 26.22% lower average root mean square error (RMSE) compared to a state-of-the-art 360-degree camera-based visual SLAM system, even without utilizing loop closure functionality, and a 39.54% lower average RMSE compared to the same system without loop closure. Moreover, this study reduced time spent on camera pose alignment by 95.80% compared to manual alignment. These results indicate that the proposed reference point–based bundle adjustment provides a practical and reproducible solution for BIM-aligned spatial analysis in construction and built-environment scenarios.