Novel Methods and Algorithms for Presenting 3D Scenes

In recent years, improvements in the acquisition and creation of 3D models gave rise to an increasing availability of 3D content and to a widening of the audience such content is created for, which brought into focus the need for effective ways to visualize and interact with it. Until recently, the task of virtual inspection of a 3D object or navigation inside a 3D scene was carried out by using human machine interaction (HMI) metaphors controlled through mouse and keyboard events. However, this interaction approach may be cumbersome for the general audience. Furthermore, the inception and spread of touch-based mobile devices, such as smartphones and tablets, redefined the interaction problem entirely, since neither mouse nor keyboards are available anymore. The problem is made even worse by the fact that these devices are typically lower power if compared to desktop machines, while high-quality rendering is a computationally intensive task. In this thesis, we present a series of novel methods for the easy presentation of 3D content both when it is already available in a digitized form and when it must be acquired from the real world by image-based techniques. In the first case, we propose a method which takes as input the 3D scene of interest and an example video, and it automatically produces a video of the input scene that resembles the given video example. In other words, our algorithm allows the user to replicate an existing video, for example, a video created by a professional animator, on a different 3D scene. In the context of image-based techniques, exploiting the inherent spatial organization of photographs taken for the 3D reconstruction of a scene, we propose an intuitive interface for the smooth stereoscopic navigation of the acquired scene providing an immersive experience without the need of a complete 3D reconstruction. Finally, we propose an interactive framework for improving low-quality 3D reconstructions obtained through image-based reconstruction algorithms. Using few strokes on the input images, the user can specify high-level geometric hints to improve incomplete or noisy reconstructions which are caused by various quite common conditions often arising for objects such as buildings, streets and numerous other human-made functional elements.