Key points are not available for this paper at this time.
Abstract. Modern measurement technologies such as Terrestrial Laser Scanning or combined Structure-from-Motion with Multi-View Stereo are commonly utilised to monitor, preserve and document cultural heritage objects and sites. For this reason, it is essential to know the capabilities and limitations of the sensor used, the data processing methods, and in particular, the orientation of the images. However, these algorithms tackle different errors and have different effects on the final accuracy of images orientation. For this reason, it is essential to know how the algorithms implemented in the Structure-from-Motion approach work. Due to the impossibility of obtaining this information for commercial solutions, it is necessary to use synthetic data to assess the quality of the SfM process. Therefore, this article aims to present the method of evaluation of SfM approach implemented in commercial Agisoft Metashape and COLLMAP open-source software based on the synthetic data generated from TLS point clouds of three different Cultural Heritage sites. In addition, obtained results were compared with the author's SfM approach based on BRISK, FAST, CenSurE, SIFT and SURF (and its Affine detectors equivalents) detector implemented (Fig. 1) and Learned-based -feature extraction approach SuperGlue and LoFTR. The second aim of this research is to propose an application to automatically generate scalable benchmark based on point clouds or 3D models of cultural heritage objects.
Markiewicz et al. (Mon,) studied this question.