Abstract:
A 2D image sequence can be a great source of motions for animating a 3D model. However, depth information cannot simply be extracted from a two-dimension image. Thus, there is a need for a method to obtain this third dimension data. Several methods have been presented over the years. But most of them employed a complex mathematical concept which makes it unavoidably slow. Moreover, there is an inherent problem of reflective ambiguity which must be addressed. In this study, we present a technique to perform a 3D coordinate estimation of 2D hand motion from an image sequence. In our method, the orthographic projection model is used to determine the Z coordination. Additionally, information from the previous frames and interdependence of a hand model are used to handle occlusion. We also propose a set of constraints on the finger joints in order to deal with reflective ambiguity. In our experiment, XY coordinates of a set of feature points are extracted from a Maya animated Hand Clinching motion. The missing data and depth information are then calculated. Finally the resulting Z coordinates are evaluated by comparing with the actual Z values from the Maya animation. The result shows that our method can estimate the Z coordinated quite well and can correctly solve the reflective ambiguity in most common cases
