A METHOD FOR ESTIMATING THE ACCURACY OF THE TIE POINTS CLOUD: THE CASE STUDY OF PALAZZO DE FALCO

Abstract

Abstract. In the context of cultural heritage documentation and conservation, an accurate 3D model is essential to properly describe an artifact of historical relevance and provide the expert user with verified metric data. In this paper we propose a workflow aimed at estimating errors in the image orientation steps, which is useful for improving the accuracy of the sparse cloud (composed of Tie Points). Some parameters provided as input to the processing software, often overlooked by the user in favour of the default ones, are analysed. Specifically, tests are carried out using the commercial software package Agisoft Metashape and involve the analysis of three parameters: (i) the accuracy of measuring the Ground Control Points (GCPs) provided by the user; (ii) the accuracy of the Tie Points (TPs) computed by the software; and (iii) the accuracy of collimating the GCPs on the image. The accuracy of the GCPs is estimated by implementing the formulas of error propagation in MATLAB, considering the three concurring sources of error: the coordinate measurement with the Total Station (TS), the coordinate transformation from the local system to the UTM/ETRF00 System, and the GNSS measurement for estimating the transformation parameters. The transformation parameters are calculated using the Bursa Wolf method. The collimation accuracy of GCPs on the image is estimated by the reprojection error of each single GCPs on the image plane. The accuracy of the GCPs will be estimated by the standard error per unit weight (SEUW), accurate GCPs being expected to have a SEUW value close to unity. SEUW is also the overall indicator for estimating the accuracy of the TPS cloud because it is influenced, in different ways, by all three parameters analysed. Test Area is a historic building located in the town of Fisciano, a few kilometres from the city of Salerno, known as Palazzo De Falco. The results highlight how these parameters, if not properly considered, can significantly affect the final 3D model.