Reconstruction and Analysis of Dynamic Shapes
Motion capture has revolutionized entertainment, sports and biomechanics, despite the fact that it tracks only a small set of points. 3D scanning methods, on the other hand, digitize complete surfaces of static objects, but are not applicable to moving shapes. I present methods that can obtain the moving geometry of dynamic shapes, such as people and clothes in motion, and then analyze it in order to advance animation. Further understanding of dynamic shapes will enable various industries to enhance virtual characters, advance robot locomotion, improve sports performance, and aid in medical rehabilitation, thus directly affecting our daily lives.
I show that a lot of the expressiveness of dynamic shapes can be efficiently recovered from silhouettes alone. Furthermore, the reconstruction quality can be greatly improved by including surface orientations (normals). In order to make reconstruction more practical, I strive to capture dynamic shapes in their natural environment, which I demonstrate with hybrid inertial and acoustic sensors. The reconstructed dynamic shapes need to be analyzed in order to enhance their utility. My methods allow animators to generate novel motions by transferring facial performances from one actor onto another using multilinear models, as well as by transferring full-body movements between different characters using a few examples. The presented methods provide some of the first and most accurate reconstructions of complex moving surfaces, and are some of the few available approaches that establish a relationship between different dynamic shapes.