In general, the Boundary Representations (B-Reps) model is used to create three dimensional models. In recent years, the spatial partitioning model has entered the spotlight as computer technology makes rapid progress. In particular, the Octree data structure has the advantage of having limited data space. It is necessary to convert Octree data into B-Reps to be able to use the data in conventional CAD systems. At present, the Marching Cubes method is a popular approach for such conversion. However, this method has the disadvantage that the number of generated polygons is increased. In this paper, we propose a new method to convert the Octree data structure into B-Reps. This method reduces the number of generated polygons and calculation time and provides high conversion accuracy. Several experimental results show that this method is effective in converting Octree data structure into B-Reps.

View full abstract

This paper presents an algorithm that generates object models using parametric surfaces from unorganized data points in 3D space. Some techniques to construct surfaces from data points have been proposed. These techniques can generate the polyhedral approximation, generating too many surfaces, or then can generate non-parametric surfaces. This paper proposes a technique of surface modeling using unorganized data points. This method proceeds as follows : the generated surface converges to a desired shape through the iteration of two processes : “control point calculation” and “reparameterization.” When the user needs a more accurate fitting, the “surface subdivision” step is performed. In the control point calculation step, the least-square method is used, and it can fit (derivative) continuous surfaces. This method can generate a surface model from data points, and it is useful for interactive surface design.

View full abstract

Representing a real-world object in a computer vision system by using a range finder is one currently important subject. However, it is difficult to display a 3-D object at high speed with raw range data because of the scattered distribution and enormous quantities of the data. An efficient method for scattered range data re-sampling by uniform cube-based segmentation is proposed in this paper. By analyzing scattered range data delivered from optical measuring devices, the outline of a 3-D object is obtained, a group of cubes are then formed by making the cross sections in object space. Then, the scattered range data are divided into cubes. Based on re-sampling of the data inside each cube, lattice points in new patches are generated. By employing the proposed method, a 3-D object can be reconstructed automatically with fewer patches and displayed in the surroundings of 3-D vision at high speed.

View full abstract

Visual simulation of seasonal mountain scenery plays a vital role in the realistic image synthesis of natural scenery. In this paper, we will present a snowfall model to simulate a close view of a snowy landscape including snowcapped garden stones or stone lanterns. Using this model, we can produce typical accumulations of snow by specifying a few parameters for the model.

View full abstract

This paper describes a modeling technique to produce both a three-dimensional map and rendering algorithms based on the digital map data of land. An algorithm for synthesizing digital map data is developed, where the transverse data for a river in two-dimensional format is transformed into three-dimensional map data. To develop a high-speed rendering technique, a method of reducing raw data without greatly affecting the image resolution is also discussed. Experiments on synthesizing a three-dimentional map and displaying it were conducted using measured river data. The results showed that the proposed modeling and rendering methods are both useful and practical.

View full abstract

In the ray tracing method, each pixel can be computed independently. Therefore, when the ray tracing algorithm is implemented on a parallel computer, a group of pixels is usually assigned to each element processor. The element processor must have all the environment data and object data for the rendering. When there are too many objects for local memory, a distorted image will be generated. In solving this problem, to date, the necessary data had to be transmitted. However, when the number of element processors is increased, efficient rendering cannot be expected because of communication overheads. This paper proposes a hybrid parallel ray tracing algorithm that combines pixel-parallel and object-parallel ray tracing. As this algorithm inherits the advantages of pixel-parallel ray tracing, high efficiency can be achieved even on a parallel computer using a large number of element processors.

View full abstract

In the field of visualization and CG (Computer Graphics), it is necessary to propose new methods to present objects precisely and to display images in real time. However, these methods need a great deal of computing time and memory. Recently, many methods, using a massive parallel computer, have been reported to satisfy these conditions. The purpose of our research is to develop practical visualization algorithms for massive parallel computers. In this paper, a parallel ray tracing algorithm is proposed, because the ray tracing algorithm has a great deal of parallelism and is able to take advantage of the high performance of massive parallel computers. The ray tracing algorithm has some parallel approaches, i.e. pixels parallel, objects parallel, rays parallel and so on. We propose a hybrid parallel ray tracing algorithm to include these parallel approaches and allowing experiments to compare processing time between each parallel approach. In conclusion, we point out problems with hybrid algorithms on massive parallel computers.

View full abstract

Emphasizing expressions are an effective way of creating motion in moving pictures. The author proposes a “Motion Filter” which is a computerized tool to design the motion of characters for animation. Computer animation has been in great demand in the field of TV processing. However most animation has been crteated by animators who paint their work directed onto celluloid in cartoon films. The writer asked those experts about their techniques and analysed these in order to be able to transfer them into a computer system. Motion Filter accepts keyframes as input motion and transforms these into emphasized motions. It works interactively and is friendly to most users who wish to design animation.

View full abstract

This paper describes a series of algorithms which makes 3D image data look like conventional cell animation images. The techniques of producing cell animation have not changed much from when movies were invented. It has essentially been a typical labor-intensive manual industry. That industry started to disappear in the 70s in the USA because of the rising cost of labor. The same thing happened in Japan in the 80s. Digital ink and paint systems were introduced to the industry in the late 80s, but the system was not yet powerfull enough to bring back the jobs that had once fled from the USA, Europe and Japan. This research started when PCs became cheaper and fast enough with a variety of 3D software packages. Making cell animation-like images by using 3D image generation software would not be practical at all if PCs were still expensive and available 3D software were slow and complicated for artists to operate.

View full abstract

In this paper, we present a realistic CG (Computer Graphics) animation method to move fishes. These have highly realistic appearance and movement. NURBS (Non-Uniform Rational B-Spline) is employed for both modeling and deformation of CG fishes. This makes the surface data of the fishes very simple to obtain. We can create the very smooth shape of the fishes (one side) by using about 8 x 4 control points of NURBS. Realistic deformation of the fish is also achieved by forming the control points into several groups and moving them. Inverse kinematics are also incorporated into this deformation. The fishes can swim naturally using a combination of 3 deformation patterns and 3 locomotion patterns. Autonomy of the fishes is realized by a mental model employing random numbers. Both schooling behaviors and collision avoidances are obtained by an algorithm employing detection vector and direction vector. Since the computing costs for this method are not high, we can run the program using general-purpose graphics workstations in realtime.

View full abstract

In this paper, we describe a system to generate CG animation as an introduction for space phenomena. We are currently able to obtain a great deal of information and data on phenomena on computer networks, but each item of information is independent and we are unable to understand the complete phenomena. Our visualization system is able to provide CG animation, using the latest information, showing phenomena which are usually invisible. This system has three functions, data integration, CG generation and interactive operation. Using these functions, we arbitrarily can observe space phenomena based on information available through the network through the interactive use of a mouse. To illustrate the system's usability, we made CG animation on the behavior of the comet Shoemaker-Levy 9 that collided with Jupiter on July 1994.

View full abstract

We have been studying a virtual camera system that generates camera images electronically in real-time based on camera operations such as panning, tilting, zooming and so on. This system is used to realize an Electronic Set that makes any camera work possible while synthesizing pictures. This paper describes a system that generates camera images by cutting out from a wide panoramic image. The main research on this includes the following.
1) Method to generate the panoramic image.
2) Method to obtain a camera image from the panoramic image.
In the first method, we generated a wide panoramic image by connecting a number of static images filmed separately by panning and tilting the TV camera. In the second method, we studied a virtual camera system to convert the panoramic image obtained to a camera image depending on panning, tilting and zooming. We will introduce experimental results on the two subjects above.

View full abstract

CG visualization of large scale city planning needs a lot of scene data, and data input has been a bottleneck in CG simulation. This paper describes an interactive scene data input system for virtual scene simulation. This system provides the functions of terrain data editing, automatic generation of road description data, design and arrangement of buildings, arrangement of trees, generation of ground surface data, and arrangement of special shape data modeled by other CAD systems. With this system, one can construct city scenery in a virtual space directly within a short time, while evaluating it.

View full abstract

We presented a simplified model for a virtual pendulum manipulation system in three-dimensional virtual space. In this system, we realized two kinds of interaction-batting interaction (hitting pendulum with racket) and KENDAMA interaction (manipulating a saucer connected to a pendulum with a string so as to put the ball on the saucer). To implement these physically-based real time interactions, we derived a simplified model for parabolic motion, the swing of a pendulum, the motion of loose string and rebounding motion. In addition, to reduce the extensive computation time to generate these physical motions. a good approximate model for each motion was developed.

View full abstract

We have developed a platform for training systems using virtual reality technologies, which are expected to be applicable in rehabilitation and the training for various sports. As a part of our research using the platform, we have implemented and installed the simulation capability of basketball dribbling, and begun a study on its effectiveness for training. The virtual reality system with force display we have developed can produce high force output allowing the control of the upper arm and forearm muscles. It is therefore able to simulate the comparatively large force needed for dribbling a rather heavy ball such as a basketball. This paper reports on the dependency of “the difficulty/easiness” of ball dribbling based on ball's size, rebound coefficient, and gravity. Measurements were conducted using our virtual environment equipped with 3 D visuals, force, and haptic interfaces.

View full abstract

We constructed a 3D game 'Tower of Hanoi' with force feedback in virtual space, using a new interface device named SPIDAR (SPace Interface Device for Artificial Reality). With the development of computer graphics, existing 3D game systems can provide us with natural virtual space. We consider that the force feedback is necessary especially when we operate 3D objects. But there is no 3D game system that can be used to express 'Picking' feelings with fingers. In this paper, we build a new type of 3D game system analyzing operations in real space. And we evaluate this system through a 'Peg-in-Hole' experiment, using the virtual 'Tower of Hanoi'.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract