Background. Industrial use of 3D reconstruction requires reconstructed models to be sufficiently reliable for measurement tasks. 

Objective. This thesis evaluates the feasibility of using smartphone-based Structure from Motion to generate scaled point clouds of Roxtec installation environments. The focus is on software-estimated measurement uncertainty and the practical use of calibration targets as scale references. 

Methods. Four case-study scenes containing Roxtec products were captured as smartphone videos by Roxtec personnel following a predefined instruction protocol. In total, nine videos were processed. Calibration targets were placed in the scenes to assign metric scale to point clouds generated using Structure from Motion. The reconstruction software reported model-specific coordinate precision, which was propagated to point-to-point distance uncertainty. Selected physical measurements were used by Roxtec to verify the practical relevance of the results. 

Results. The estimated 95% point-to-point measurement uncertainty ranged from 0.16 mm to 0.45 mm across the case-study reconstructions, with point densities ranging from approximately 3 to 27 points/mm2. The scenes were captured at approximate camera distances of 0.5–0.8 m. 

Conclusions. The results indicate that smartphone-based Structure from Motion can produce scaled point clouds with sub-millimeter estimated measurement uncertainty for the studied scenes. However, reliability varies locally within each reconstruction. Scene-average metrics are therefore insufficient on their own and local reconstruction quality must also be considered when deciding whether measurements from a point cloud are dependable. 

Keywords. Structure from Motion, photogrammetry, point cloud, reconstruction error, measurement uncertainty.