ABSTRACT
Recent progress in shape modeling allows the recovery of precise shape models from visual data, such as color and depth images. These models can come in different representations, usually points, surfaces, or volumes, depending on the strategy employed for their estimation, e.g., stereo vision or sensor fusion, which were discussed in preceding chapters. When considering visual information over time, as in videos, temporal sequences of models can be obtained. Yet, whatever the representation, static reconstructions performed independently over time do not provide clues on the motion and the deformation of a shape. However, natural scenes are usually dynamic and composed of shapes that evolve over time, for instance humans and objects (Figure 12.1). Several applications are based on the capture of shape evolution over time: in digital content production, for instance, for realistic animation of virtual characters; or in motion analysis for sport or medical applications. The next step in modeling reality is therefore concerned with the ability to recover or capture motion and deformation of shapes of unknown types by using visual information sampled over time.
