Innovative methodologies for multi-view 3D reconstruction of cultural heritage

Abstract

This dissertation focuses on the use of multi-view 3D reconstruction techniques in the field of cultural heritage. To name just a few applications, a digital 3D acquisition can be used for documentation purposes in the event of destruction or damage of an artefact, or for the creation of museums and virtual tourism, education, structural studies, restoration, etc... All these applications require high precision and accuracy to reproduce the details, but there are other important characteristics such as low cost, ease of use, the level of knowledge needed to operate the systems, which have also to be taken into account. At the present time, the interest is growing around the use of images for the digital documentation of cultural heritage, because it is possible to obtain a 3D model by the means of common photographic equipment. In this work, we have investigated multi-view 3D reconstruction techniques in two specific fields that have not been treated in literature: the 3D reconstruction of small objects (from few mm to few cm) and the survey of submerged archaeological finds. As for the 3D reconstruction of small objects, a new methodology based on multi-view and image fusion techniques has been developed. The used approach solves the problems related to the use of macro lenses in photogrammetry, such as the very small depth of field and the loss of quality due to diffraction. Since image matching algorithms cannot work on blurred areas, each image of the sequence is obtained by merging pictures acquired at different focus planes. The methodology has been applied on different case studies, and the results have shown that it is possible to reconstruct small complex objects with a resolution of 20 microns and an accuracy of 10 microns. For which concerns the underwater imaging, a preliminary comparative study between active and passive techniques in turbid water has been conducted. The experimental setup consists in a 3D scanner designed for underwater survey, composed by two cameras and a projector. An analysis on the influence of the colour channel has been conducted, showing how it is possible to obtain a cleaner reconstruction by using the green channel only. The results have shown a denser point cloud when using the passive technique, characterized by missing areas since the technique is more sensible to turbidity. By contrast, the reconstruction conducted with the active technique have shown more stable results as the turbidity increases, but a greater noise. A multi-view passive technique has been experimented for the survey of a submerged structure located at a depth of 5 meters, on a seabed characterized by poor visibility conditions and the presence of marine flora and fauna. We performed an analysis of the performances of a multi-view technique commonly used in air in the first instance, highlighting the limits of the current techniques in underwater environment. In such conditions, in fact, it has not been possible to obtain a complete reconstruction of the scene. The second stage of the process was the testing of image enhancement algorithms in order to improve matching performances in poor visibility conditions. In particular, a variational analysis of the factors that influence the quality of the 3D reconstruction, such as the image resolution and the colour channel, has been performed. For this purpose, the data related to the parameters of interest, such as the number of features extracted or the number of oriented cameras, have been evaluated. The statistical analysis has allowed to find the best combination of factors for a complete and accurate 3D reconstruction of the submerged scenario.