In CG animation, the method for modeling of CG characters such as humans or animals is important. If these shapes are represented by parametric surfaces, such as Bezier, B-spline, or NURBS surfaces, complex procedures are required to make adjacent surfaces connect smoothly. To solve this problem, instead of parametric surfaces, subdivision surfaces are often used to represent shapes. However, how best to fulfill efficient modeling intended by designers is still problematic. Another method is proposed to model complex freeform surface shape, with which the boundary curves of a surface are specified to create a curved mesh and mesh area within the boundary is interpolated with Gregory patches. This kind of modeling method enables to modify a curved mesh directly with intuition so as to achieve a desired shape. In this paper, to take an advantage of the features of both Catmull-Clark subdivision surface and Gregory patch, we propose a local modification method that generates a curved mesh from a polygon mesh with the same topology using a Catmull-Clark subdivision surface, modifies the generated curved mesh directly and applies the modified curved mesh shape to the polygon mesh. One of the characteristics of our method is that modification of some vertices or edges affects only their adjacent surfaces. Also, the surfaces within a specified range can be modified freely with maintaining the continuity between the surfaces and their adjacent surfaces outside the specified area. Applying modification of a curved mesh to a polygon mesh enables local modification or intuitive modification that are rather difficult to perform for a polygon mesh.

View full abstract

In this paper, we propose a new calligraphy drawing system based on the principles behind the electronic music sampler. Our purpose is to render the traditional expression of calligraphy brushstrokes digitally and to propose a new mode of expression by taking after the sampler metaphor. By this system, the user scans over reference calligraphy images by tracing the stroke with a mouse-pointer, and rewrites a calligraphy by outputting sampled images on the drawing window. We designed the system with the idea of a music sampler device in mind, and built on this metaphor by adding further editing functions; envelope generator and mixer. Moreover, we installed a function that allows referencing back to the original images. This option allows us to relive, in graphic terms, the pleasure of discovering the original work behind a piece of sampled music.

View full abstract

Recently, many interactive systems that talk with CG characters using voice recognition are proposed. However, in the traditional interactive systems, there is a problem that the interval of conversation is unnatural since processing of voice recognition is slow. In this paper, therefore, we propose a technique for generating reactive attention based on user's nonverbal information and a conversation status.In order to realize our technique, we first get verbal data and 3 nonverbal data of a user. And what CG characters pay attention to is decided on those parameters. Then, attention of CG characters is generated by attention angle shearing mechanism. An example of a conversation with CG character using our technique demonstrates the effectiveness of adding attention of CG characters to interactive systems.

View full abstract

This paper presents a video content retrieval system of operators' performance areas for camera monitoring and life-logging on a plant operator, fine artist etc. As the system focuses on grasping actions and tool handling operations, it records not only audio-video data, but also grasping pressures by special finger pressure sensors. A sensing module consists of tiny film sensors and a bluetooth wireless data transmitter, in order not to interfere workers' free actions. Unlike other video retrieval system with voice or image recognition, tag text analysis, the proposed system searches the desired video scene based on grasping pressure data, so that it distinguishes whether the worker operates a tool physically while the worker's outlook shows holding. As its content browser places only interested video clips on a time line, which the worker handles the tool with a certain force, the user easily knows actual actions among a longer and monotonous raw video sequence. By applying a supervised learning process with known actions in NN-method, the system extracts the same actions from a video sequence when the user tells the representative action as a search key in the content browser.

View full abstract