3D Computer Vision aims at enabling computers sense, model and analyse the 3D information of the world around them. During the last decades, it has made impressive advances and nowadays is behind many real-world applications, such as visual effects in films, augmented reality, natural human-computer interaction, body motion capture for video gaming and 3D imagery for web mapping, to name a few. However, today's 3D computer vision systems have still several important limitations, since they can only work reliably under restrictive conditions.

In this talk, I will present my recent and ongoing research efforts towards developing methodologies that overcome these limitations, leading to robust systems that work under almost any condition and use low-cost acquisition devices. I will discuss how our dense variational formulations succeed in performing accurate dynamic 3D reconstruction of deformable objects, using as only input the video from a single camera, without requiring any markers or additional sensors. Apart from the case of generic deformable objects, I will also present our methods for human faces specifically, where building high-quality, large-scale 3D shape models and reliable algorithms to fit them on image data is of paramount importance. Finally, I will talk about applications of our dense 3D face modelling to corrective craniofacial surgery on patients with facial conditions.