PHOTOGRAMMETRIC AUTOMATION: IS IT WORTH?

Abstract

This paper focuses on the evaluation of automated image based techniques which are used lately in order to produce three dimensional digital models of cultural heritage objects. This implies a doubt as to how accurate and reliable are the products of these automated algorithms and how efficient they are for providing the necessary 3D material for the virtual environments. The implementation of this innovative approach has gradually become very popular in the field of Cultural Heritage during the last five years. Non-specialists found a way to easily produce 3D reconstructions just by taking a few images and using Structure-from-Motion and Multiview Stereo algorithms implemented by commercial or open source software. However, this fact led to debatable results, as a lot of ambiguities are lurking hidden in the “happiness” of automation. In order to assess the metric performance of these algorithms an innovative metric evaluation strategy has been designed. A very accurately measured test field, set up mainly for calibrating digital cameras, was used as an object of well-known ground truth. It was established that the accuracy of the resulting 3D reconstruction depends on the spatial analysis and the general quality of the original digital images, on the careful selection of the parameters provided by the software, the properties of the object itself and the computing power available.