ムバラキ ハッサン,田中弘美,北村泰一
Evaluation of the Homotopy Sweep Technique Representation and Animation of
three-dimensional images for human face generation and mouth animation.
Abstract:The purpose of my internship was to evaluate an interpolation method based on the homotopy between two arcs and to apply it to the representation and description of three dimensional data taken from images in both a static and a dynamic approach. My work is related to the field of human representation in the ATR Artificial Intelligence Department and fits into the global project of the new teleconferencing system with realistic sensations in a real time process
This real time process requires a special processing because real time computer data representation and animation are limited by computers' and neworks' capabilities. Even though many objects may be constructed by assembling geometric shapes like cubes, spheres, or pyramids, most objects have a free-form shape like human faces. One can use the coordinates of the very high number of points given by a 3-d digitizer but this is inconvenient in the case of a computer finite storage at the prospect of a data transmission for a real time animation. We need methods for developping free-form surfaces easily from a limited set of data and a model to allow people to visualize and understand the structure of the modeled entity. Furthermore, this model should provide a convenient vehicle for experimentation with.
The Homotopy Sweep Method was proposed in 1991 for surface generation using a set of two-dimensional contours with the interesting ability to control the transition from one contour to the other. During the five months of my internship in ATR, I was given the task to evaluate the Homotopy Sweep Technique for human face representation and to implement it on Silicon Graphics Workstations using three-dimensional data given by a digitizer. At the same time, we were thinking about a generalization of this method and, in order to take advantage of the convenient control of the deformation, I developped another interpolation formulation based on a dynamic approach. With some assumptions concerning the muscles and the human morphology, I applied this dynamic approach to mouth animation using a Lip Tracking Process and data given by a three-dimensional digitizer. With this method, we can control mouth continuous deformation just by specifying a few parameters. The animation results (recorded on VHS tape) were very realistic and the simple interpolation computations can be considered as a real time process.
In this report, I introduced the Homotopy Sweep Technique and the differents problems I had to solve for its application in the two approaches. Section 2 contains the basic principles, notions and definitions related to the Homotopy Sweep Technique. The static approach and the assumptions of our implementation are developped in Section 3. Section 4 and the VHS tape show the
diferents aspects of the dynamic approach for the human mouth animation and the powerful ability to control the deformation.