The critical evaluation of Multi-View Stereo approaches in 3D dense reconstruction for Cultural Heritage objects

Abstract

The article aimed to analyse the possibilities of using Multi-View Stereo algorithms implemented in commercial (Agisoft Metashape and Pix4D) and open-source (OpenMVS 2.0) solutions. To critically assess the correctness of 3D shape mapping, the following parameters were analysed: density of point cloud, planarity and roughness distribution. The analysis used photos obtained from three different types of sensors, i.e. the Hasselblad H6d-100 medium format camera (fragment of frescoes - Test Site I), low-cost Olympus C-5050Z (fragment of archaeological excavations - Test Site II) and Canon Mark II (fragment of a historic room). The analyses performed show that the best results were obtained for point clouds generated using the Agisoft Metashape software and worse for Pix4D. Evaluating the results obtained for the OpenMVS, it can be noted that better quality in terms of geometric accuracy was obtained for the patch-based algorithm than for the semi-global matching algorithm. The choice between Agisoft Metashape and OpenMVS2.0 depends on the user's knowledge and experience in programming and photogrammetry, as OpenMVS does not provide a user interface but provides an open-source alternative to Metashape.