Uses of uncalibrated images to enrich 3D models information

The decrease in costs of semi-professional digital cameras has led to the possibility for everyone to acquire a very detailed description of a scene in a very short time. Unfortunately, the interpretation of the images is usually quite hard, due to the amount of data and the lack of robust and generic image analysis methods. Nevertheless, if a geometric description of the depicted scene is available, it gets much easier to extract information from 2D data. This information can be used to enrich the quality of the 3D data in several ways. In this thesis, several uses of sets of unregistered images for the enrichment of 3D models are shown. In particular, two possible fields of application are presented: the color acquisition, projection and visualization and the geometry modification. Regarding color management, several practical and cheap solutions to overcome the main issues in this field are presented. Moreover, some real applications, mainly related to Cultural Heritage, show that provided methods are robust and effective. In the context of geometry modification, two approaches are presented to modify already existing 3D models. In the first one, information extracted from images is used to deform a dummy model to obtain accurate 3D head models, used for simulation in the context of three-dimensional audio rendering. The second approach presents a method to fill holes in 3D models, with the use of registered images depicting a pattern projected on the real object. Finally, some useful indications about the possible future work in all the presented fields are given, in order to delineate the developments of this promising direction of research.