Document Type : Original Research Paper
Authors
Dept. of Geomatics, Faculty of Civil Engineering, Babol Noshirvani University of Technology
Abstract
Background and Objectives: Large maritime structures increasingly require dense, geometrically reliable 3D documentation for dimensional control, maintenance planning, reverse engineering, and digital-twin workflows. Close-range photogrammetry is attractive for such tasks because it can generate dense 3D products from ordinary cameras with lower equipment cost and greater operational flexibility than many conventional dense-capture techniques. Nevertheless, the practical performance of photogrammetric processing software is not necessarily identical, especially when the object is very large and contains reflective metal, low-texture painted surfaces, sharp structural edges, and curved hull geometry. The objective of this study was to compare three widely used commercial photogrammetric software packages—Metashape, Pix4DMapper, and ContextCapture—under one industrial-scale case study and to determine how their outputs differ in geometric accuracy, local surface agreement, reconstruction completeness, processing time, self-calibration behavior, and practical usability.
Methods: Images were acquired with a non-metric iPhone smartphone camera using automatic focus and exposure. After five low-quality images were excluded from an initial set of 455 photographs, 450 images were retained for processing, including 288 approximately parallel images and 162 convergent images. Fifty-two reference targets were distributed over the visible side of the vessel and measured by industrial surveying with a Leica TS09PLUS total station. Thirty targets were used as control points and 22 targets were reserved as independent check points. The same image set and reference coordinates were processed in Metashape, Pix4DMapper, and ContextCapture. Evaluation included the root mean square error of control and check points in the three coordinate components and in three dimensions, visual and numerical assessment of point-cloud completeness, cloud-to-cloud distance analysis in three representative zones with edged, planar, and curved geometry, processing time for alignment and dense reconstruction, interpretation of self-calibration outputs, and an observational usability assessment.
Findings: The independent check-point results showed the largest separation among the software packages. The 3D RMSE was about 5 mm for Metashape, 40 mm for ContextCapture, and 45 mm for Pix4DMapper. The component-wise errors of ContextCapture and Pix4DMapper were concentrated mainly in the horizontal components, while the vertical component was much smaller. In the local cloud comparison, ContextCapture showed consistently smaller distances from the Metashape reference cloud than Pix4DMapper. For the edged, planar, and curved zones, the root mean square cloud-to-cloud distances were 22, 24, and 20 mm for ContextCapture and 33, 41, and 37 mm for Pix4DMapper, respectively. The planar zone produced the largest separation between the two compared point clouds. Visual inspection also indicated a denser and more complete reconstruction by ContextCapture in several geometrically difficult regions, whereas Pix4DMapper exhibited missing or less complete point-cloud coverage in some areas. Under the reported test configuration, total processing times were 45 min for Metashape, 67 min for ContextCapture, and 86 min for Pix4DMapper. The software packages also differed in workflow complexity and the level of user control over processing parameters.
Conclusion: The results demonstrate that software choice can materially affect the geometric and operational outcome of close-range photogrammetric modeling of a large ship hull even when the input imagery and reference information are held constant. In this data set, Metashape produced the lowest independent check-point error and the shortest recorded processing time, while ContextCapture produced the closest local agreement with the Metashape reference point cloud and favorable surface completeness. Pix4DMapper offered a comparatively simple workflow but produced larger independent check-point errors and larger local cloud distances in the evaluated zones.
Keywords
Main Subjects
COPYRIGHTS
© 2026 The Author(s). This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
)