Assessing the Metric Accuracy of Free iPhone LiDAR Applications for Multiscale 3D Documentation

Abstract The integration of LiDAR sensors into consumer smartphones has opened new possibilities for low-cost 3D documentation, yet the metric accuracy of free iPhone LiDAR applications across different spatial scales and environmental conditions has not been rigorously characterized. This study benchmarks three freely available applications (Modelar, 3D Scanner, and SiteScape) against a high-precision Trimble X7 Terrestrial Laser Scanner (TLS) reference across three case studies at the University of Debrecen: an outdoor building façade (46.5 m), a GNSS-denied indoor corridor, and a small-scale bronze statue. The façade and corridor datasets were georeferenced using a total-station control network, whereas the statue used an independent local reference frame. Point clouds were aligned to reference data using control-point registration followed by iterative closest point (ICP) fine registration, and surface deviations were quantified using multiscale model-to-model cloud comparison (M3C2). Geodetic accuracy follows standard smartphone-LiDAR evaluation protocols, with pre-ICP alignment quantifying fine registration contribution as a 32–57% RMS reduction at the façade depending on application. Results revealed a consistent scale-dependent pattern, with RMS deviations of 1.3 to 2.0 cm for the statue, 8.1 to 10.3 cm for the corridor, and 3.7 to 8.5 cm for the façade (without and with glazed openings, respectively). Excluding window and entrance points reduced M3C2 standard deviations by up to 50%, indicating that specular reflection from glass, rather than generic SLAM drift, is the dominant source of façade-level deviation. SiteScape experienced a complete loss of trajectory consistency in the corridor, manifesting as a 180° wall inversion, showing that algorithm behaviour, not raw point output, determines usability in long, feature-poor, GNSS-denied spaces. Modelar produced the fewest points yet achieved the lowest deviations across all three case studies, indicating that point density is an unreliable proxy for geometric quality. Together, these findings offer a geodetically controlled benchmark informing practitioner decisions when selecting free iPhone LiDAR applications for heritage recording, structural inspection, or early-stage BIM workflows, while underscoring current limits relative to TLS.