Abstract

Reconstructing the 3D shape of an object from images is a longstanding challenge in computer vision. While for rigid objects there are stable methods that have been used in practical applications, the reconstruction of deformable objects is still a widely open problem.

In this talk, I will discuss how an inextensibility prior can be used to better constrain the highly ambiguous problem of reconstruction of deformable objects from images or videos taken with a single camera. I’ll present two new techniques that make use of the inextensibility prior: (1) a unifying approach that can be applied to two tasks that have been so far considered independently – template based reconstruction and non-rigid structure from motion – producing comparable or better results than the state-of-the art methods and (2) a template based method designed to reconstruct closed surfaces, making it applicable to the reconstruction of objects with volume, such as animals, that have not been previously attempted with existing template methods.