Lower Cost Spatially Immersive Visualization for Human Environments

 

 

Frederic I. Parke, Ph.D.

 

Visualization Sciences

College of Architecture

Texas A&M University

College Station, Texas, USA, 77843-3137

 

Phone: (+1.979.845.3465)

Fax:     (+1.979.845.4491)

E-mail: parke@viz.tamu.edu

 


Professor Frederic I. Parke

Visualization Sciences

216A Langford, 3137 TAMU

College of Architecture

Texas A&M University

College Station, Texas, USA, 77843-3137

 

 

Abstract

 

Access to computer simulation and computer based visualization has dramatically impacted our ability to design complex human environments.  Immersive visualization, with its ability to present high quality interactive three-dimensional representations of environments, is the next step in this evolution.  This paper explores the development of lower cost modular immersive visualization systems as a way to extend and augment our ability to plan, design and evaluate human environments.

 

Technology now available enables spatially immersive visualization systems created using off the shelf components including high performance, relatively inexpensive, commodity computers, inexpensive commodity projectors an open source software.  Flexible modular configurations utilizing polyhedral display surfaces with many identical modular components and networked visual computer clusters is one approach to such systems. 

 

Work is underway at the Texas A&M College of Architecture focused on developing and evaluating several prototypes of this new class of systems to determine their practicality and effectiveness. Underlying concepts, issues and trade-offs related to the design and development of these systems are presented.  Initial applications using these systems in human environment planning, design and evaluation are discussed.

 

 

Keywords – modular visualization systems, spatially  immersive visualization, visualizing human environments, virtual environments, parallel visual computing.

 

Introduction

 

This paper explores the characteristics and potential of lower cost modular immersive visualization systems.  We see the development and use of these systems as one way to extend and augment our ability to plan, design and evaluate human environments.

 

Computer simulation and computer based visualization are now common techniques that we use in the planning, design and evaluation of human environments.  Access to these methods has dramatically impacted our ability to deal with these complex environments.  Immersive visualization, which allows us to explore and interact with high quality three-dimensional representations of environments, is the next step in this evolution. While standard visualization techniques provide ‘windows’ into virtual environments, immersive visualization provides the sense of being ‘within’ and experiencing these environments.

 

Immersive visualization has, until quite recently, been associated with very expensive specialized systems used in applications such as scientific visualization, flight training and petroleum exploration where the benefits justified the expense.

 

Impediments to the broad use of immersive visualization have been

 

1)     high initial system cost – these systems, both hardware and software, have been expensive

2)     high cost of operation – specialist support staff and ongoing maintenance are required

3)     accessibility – only a few systems are in place for a relative small number of users

4)     software complexity – there are only a few ‘off-the-shelf’ applications, custom application software development is required for most new applications

5)     ease of use  issues – special effort is needed to use these systems, they are not well integrated into workflows except for a few specialized problem domains

6)     human factors issues – user fatigue, ‘simulator sickness,’ and the need to wear special viewing apparatus are a few of these issues.

 

One goal, aimed at mitigating high system costs and accessibility, has been to develop very capable lower cost immersive visualization systems that are useful, cost effective and widely accessible.  Such lower cost systems promise to enable much broader use in many disciplines including the development and evaluation of human environments.

 

Immersive visualization systems fall into two categories - head mounted displays (HMDs) and spatially immersive displays (SIDs) such as the CAVE concept developed at the University of Illinois at Chicago (Cruz-Neira, 1993).  Head mounted displays are designed to present two separate views of the virtual environment to the user - one view for the right eye and another view for the left eye. These two stereo views are fused by the viewer's perceptual system in the same way as right and left eye views of natural scenes.  These systems have typically required the user to wear a helmet-like apparatus that holds the left and right eye display devices.  HUD systems must, in addition to computing the synthetic views, track the viewer's location and viewing orientation in the virtual world to create the correct right and left eye views (3Space, 1987).

 

Specific challenges for HUD systems include the need for fast, accurate sensing of user head position and orientation and the need for rapid image updates with minimum latency between head motions and the corresponding image update.  If the displayed images lag behind head motions, user disorientation and a form of motion sickness can result. Another challenge has been to develop miniature high-resolution display devices and display optics that allow natural wide field of view images.

 

In this discussion we are focused on spatially immersive systems. In fully immersive systems, the viewer is surrounded by projected images. In CAVE installations, for example, the immersive environment consists of multiple large rear-projected display screens that form the walls of a presentation environment cube similar to a small room. Fully immersive CAVEs have six planar image surfaces forming the walls, floor, and ceiling of the environment. Each image presents a large portion of the possible field of view.

 

In addition, there are semi-immersive systems that provide partially surrounding images.  In one form, these make use of a large curved screen or wall that provides a wrap-around image with a wide horizontal field of view – up to 180 degrees.  These systems support significant horizontal peripheral vision, a limited vertical field of view but no rear images.  An alternative form is to use two or more abutted flat screens to approximate the curved screen.

 

Most CAVE installations are in fact semi-immersive; having fewer than the maximum six display surfaces. Many use only three display surfaces - front, left, and right - and three image projectors.  A fourth display surface and a fourth projector utilizing a reflecting mirror is sometimes added for the floor or ceiling.

 

Specific challenges of immersive systems include the need for relatively large, usually high ceiling spaces to accommodate the display surfaces and placement of the display projectors.  For fully immersive systems there is usually the need to suspend the system above the actual floor to accommodate projection onto the lower display surfaces.

 

Major Components

 

A spatially immersive visualization system consists of three major elements; the computa­tional infrastructure, the surrounding display surfaces and projectors, and the viewer tracking and interaction devices. We are exploring new approaches to both the computational infrastructure and the display surface geometries used. 

 

Until recently, the computational infrastructure of these systems was the province of specialized graphics supercomputers and proprietary software.  These systems were specifically configured to meet the very demanding requirements of immersive visualization.  These requirements include rapidly presenting multiple synchronized high-resolution images based on large sets of complex three-dimensional data.  For some applications, these images need to be updated at rates of up to 60 new images per second. These systems also need to support interactive manipulation of the three-dimensional environment data.

 

Spatially immersive systems make use of projected images.  The images may either be front projected or rear projected.  Some semi-immersive systems use front projected images.  Many immersive systems make use of rear-projected images.  Until recently the projected images have been created using expensive projectors driven from specialized, expensive, high-end graphics systems.

 

Since the display surfaces of these systems have often been curved and often required blending multiple projected images, expensive high light output CRT based projectors coupled with specialized optical image blending techniques have been the norm.  These projectors allow electronic warping of the projected images to compensate for various optical distortions. Also, the stereo imaging techniques usually employed required very high frame rates that only these expensive projection systems could support.

 

Since the primary reason for using immersive systems is to explore and manipulate elements within virtual environments, special three-dimensional position and orientation measuring devices and supporting software are used.  In addition to user position tracking, these allow users to navigate within the environments and to select and manipulate elements of the displayed environments.

 

Stereo Presentation

 

In spatially immersive systems, as in the head mounted display systems, true stereographic images are often desired.  These have usually been presented as time sequential stereo images.  The right eye and left eye images are projected alternately at a high frame rate. The viewer is required to wear special glasses that have liquid crystal shutters whose operation is synchronized to the display frame rate.  The shutter glasses allow the right eye to only see right eye images and the left eye to see only left eye images. The human visual system integrates these views into perceived three-dimensional views. As with the head mounted display version, the position and perhaps orientation of the viewer must be tracked if correct stereo views are to be projected. For systems with multiple projected images, the update rate and frame rate of all the images must be carefully synchronized.

 

An alternative stereo display approach is one that uses two projectors for each display screen – one for the right eye view and one for the left eye view.  Polarized filters are placed in front of each projector lens. The projectors are paired with filters of opposite polarizations. The viewer also wears glasses with correspondingly polarized filters.  These viewing filters are oriented so that the right eye sees only the right eye image while the left eye sees only the left eye image.

 

Yet another approach, which we have found to be quite effective, is to separate the right and left eye views based on color.  Most computed images have three color channels – red, green and blue. In the anaglyphic color based approach, only the red color channel is computed for the right eye view while the green and blue channels are computed for the left eye view. When viewed through red and cyan filters, the right eye sees the right image and the left eye sees the left images.  Amazingly, the human visual system does a good job of merging these views into a perceived three-dimensional representation.  An advantage of this approach is that the two views coexist in the same graphics image and requires only one projector per display surface.

 

 

 

 

Modular Systems

 

Current and near future technologies and computational economics allow the development of better and more cost effective spatially immersive visualization systems.  In recent years, low cost commodity projectors have been replacing the expensive projectors and commodity PC based graphics systems have been replacing expensive graphics system.

 

A very compelling concept is collections or clusters of commodity computers networked to form powerful inexpensive distributed parallel com­puting engines. Implementations of this concept often make use of extensions to the Linux operating system (Hekman, 1997).  One example of this is the Beowolf cluster concept (Sterling, 1999).

 

This very compelling concept has been extended into visual computing with the development ‘tiled’ display systems, primarily through work done at Princeton, Stanford and the Argonne National Laboratory (Hereld, 2000). Tiled displays are formed by dividing a two-dimensional display area into an array of adjacent regions or tiles. Each of these regions is projected by one of an array of image projectors. These systems make use of multiple computers, some with commodity graphics cards to drive the projector array, organized by software such as WireGL and its successor Chromium, to support large, very high aggregate resolution displays (Humphreys, 2001)(Humphreys 2002).

 

Based on published performance benchmark results, collections of relatively low cost commodity visual systems compare very favorably in both cost and aggregate performance with the expensive high-end graphics systems that have typically been used to support immersive systems. In recent years the advancement in commodity graphics workstation performance has been phenomenal (Viewperf, 2002). From a purely raw performance viewpoint the former high-end systems have clearly been overshadowed. The high-end systems do, however, still hold some advantages in terms of internal data bandwidth, the ability to handle extremely large image data sets and their ability to carefully synchronize multiple display images.

 

Having the visual computing distributed over a collection of processors allows innovation in the structure of the display surfaces.  The aggregate display surface may be composed of many planar display faces or facets.  In such configurations, each facet need only display a relatively small portion of the total virtual environment. The graphics computation needed for each display facet falls within the capacity of today's, and certainly tomorrow's, high-end commodity graphics systems.

 

Spatially immersive systems are created by arranging the display tiles or facets into a surrounding three-dimensional display surface and creating a commodity based computational architecture optimized to support such fully immersive systems. The computational infrastructure used is, as in the tiled display concept, a visual computing extension of the commodity computer cluster concept. 

 

In recent years, several open source software development environments have been created to support the development of immersive visualization systems.  These include VR Juggler (Bierbaum, 2001) (VR Juggler, 2004), Syzygy (Schaeffer, 2003), and OpenSG (OpenSG, 2004).  These are not application packages, but rather, provide the basic utility services and system support software upon which specific applications can be built.

 

The Ideal System

 

It can be argued that the ideal spatially immersive environment would be one where the user is surrounded by a seamless spherical display surface that provides very high resolution, high update rate, 360-degree panoramic stereo views of extremely complex data.  Spherical domed display surfaces, requiring specialized optics, have been used for many years in flight training simulators (Reno, 1989) and dodecahedron approximations to spherical projections have been developed (McCutchen, 1991). 

 

One concept has been to develop immersive visualization systems utilizing modular polyhedral display surface structures that are good approximations to the ideal sphere.  In this case, a high performance commodity graphics processor is included in each of the computational nodes.  There is at least one computational node for each display facet. The result is a powerful spatially immersive visualization system.

 

The faceted display elements may be arranged in a number of possible configurations.  In fact, the display facet configurations can be tailored to meet specific application needs.  There are a number of polyhedral configurations whose faceted surfaces are good approximations to the ideal spherical display surface.  Several of these are formed using many instances of a single planar shape. These polyhedra require from 12 up to 60 or more planar faces (Holden, 1971) (Wenninger, 1971).

 

Examples of such systems include the GarnetVision prototype developed by Iwata (Iwara, 1996) which used a 12 facet rhombic dodecahedron to form a fully immersive system.  The image quality and image resolution of this prototype was limited since only one graphics computer was used to created multiple image windows that were distributed to the facets using video techniques.

 

Another, commercial system, has been developed by Link Simulation Systems (Dugdale, 1999).  This system, called SimuSphere, is based on a pentagonal dodecahedron facet structure.  Systems using 3, 5, 7, or 9 of the 12 possible display facets, have been incorporated into fighter aircraft training simulators.  In these systems, the display images are computed using the graphics supercomputer approach.

 

Our approach is focused on the 24 facet Trapezoidal Icositetrahedra (Parke, 2002). An illustration of how such a 24 facet system might be located within a high ceiling space is shown in Figure 1.  The human figure shows the scale of this 5 meter diameter display structure. The building entrance shown is approximately 6 meters wide. The array of projectors needed for such a system is illustrated in Figure 2.  Each of these projectors is driven from its own graphics computational node.  The set of these 24 network connected computational nodes would form a visually extended cluster system as discussed above.  A simulated cross-section view through this proposed 24 facet structure is shown in Figure 3.  The human figure is included to provide scale and to show where the viewer would be located relative to the display surfaces.  Simulated molecular images are shown projected on the display surfaces.

 

Development Objectives

 

Our overarching goal is to help enable the effective use of immersive visualization in a much wider range of applications and disciplines.  The development of effective lower cost systems is key to this goal.  We are concerned with determining whether these visualization environments are practical and effective.  The objectives in this work can be summarized as follows:

 

1)     Explore and evaluate, at the conceptual level, possible geometric display structures and their implications for modular immersive environments.

2)     Construct, apply, and evaluate operational prototype systems for the most promising configurations.

3)     Develop or adapt the software and software libraries needed to support the operational prototypes. Some of the issues to be addressed include; effective distribution of the graphic computation, data formats, dynamic data partitioning, synchronization of the displays, the required non-standard image projection algorithms, user interaction and image compensation. Image compensation is required because of the image distortions inherent in commodity projectors.

4)     Develop applications to test and evaluate these systems.

5)     Develop valid experimental designs to measure the effectiveness of the systems.

6)     Develop a deeper understanding of the technical trade-offs and effectiveness issues of these systems.

7)     Investigate the feasibility of mass replicated, modular implementations of these systems.

 

In addition to the design, development and application of these prototypes systems, we are concerned with the effectiveness of the immerse experiences they will provide. That is, do they empower the participant by affording new insights and deeper understandings, as well as facilitating performance generally? Evaluation includes both technical evaluations of the systems and effectiveness evaluations. Technical evaluation addresses issues such as computational complexity, computational loading, dynamic performance, and cost/performance measures.

 

Results to Date

 

We are currently in the operational prototype and initial application development phases of this project. Development of the prototypes has involved detailed simulation, physical structure design, physical fabrication and assembly, supporting software design, and software implementation.

 

We selected for our initial immersive display surface structure the 24 facet polyhedra described above and illustrated in Figures 2.  To gain a good understanding of this structure, a series of increasingly larger scale structure prototypes were constructed.  The initial versions were small-scale paper and cardboard models.  Next was a one-quarter scale wood frame model.  This was followed by a three-quarter scale prototype with fabric projection screens.  This non-operational 3.5 meter diameter prototype is shown in Figure 4.  A full-scale surrounding display structure for this configuration is expected to be about 5 meters in diameter. 

 

The structural members are milled wood with a specific cross section that allows the display screen facets, which are interior to the frame, to fit together with minimal seams. The frame members also act as partial light baffles between the adjacent screens. While specialized expensive screen materials are available, we simply used fabric screens in our development prototypes.  We have found that a densely woven synthetic fabric such as white taffeta works quite well as a rear projection screen material.  This fabric would not work if the polarized stereo approach were used.  Special polarization preserving screen materials would be required

 

Standard visualization software generally assumes single rectangular display screens driven from single computers. The software we have developed allows the images to be computed for display on an arbitrary collection of convex polygonal screens. The configuration of the screens is specified in run-time files that can be easily changed as required. Since in these systems each screen facet is driven from its own computational node, the software must coordinate and synchronize the operation of multiple network connected computers. The software must also enable interactive control of the system and distribute interactive control across the network of computers in real-time. This software has been developed for the Linux operating system (Hekman, 1997) using standard network protocols and the OpenGL graphics environment (Woo, 1997).

 

We are using 1024 x 768 resolution commodity LCD projectors.  These projectors have fixed image geometry and optics that are optimized for a completely different application.  The relatively inexpensive projection optics introduce some non-linear distortion into the projected images.  These projectors, unlike their more expensive CRT counterparts, do not allow electronic image warping.  The computed images must be computationally pre-warped to compensate for projection distortions.  Raskar, et al. (Raskar, 1999) outlines an approach for doing this that we have adapted.

 

Commodity projectors also have significant light output variation across the images they project.  It is possible to compensate for this variation by computationally adjusting brightness across the images.  We have not found this necessary for our prototype systems.

 

To date we have developed two operational prototypes.  One is a three facet section of the 24 facet polyhedra.  It is shown in its inactive state in Figure 5 and in operation in Figure 9.  There are actually four facets in this structure, but only 3 are active.  The other prototype, shown in Figure 6, also has three display facets.  In this case the facets are a section of a 10 facet surrounding cylindrical structure.  These three facets provide a 108 degree horizontal field of view.  Five facets would provide a 180 degree field of view.

 

System Costs

 

The hardware cost to replicate either of these three facet systems would now be about $22,000.  This corresponds to a cost per facet of about $6,500 plus the cost of network connectivity, input control devices, and miscellaneous peripheral costs.  The per facet cost breaks down as about $3,000 for the high performance graphics computer, $3,000 for the projector, and $500 for the display screen and support structure.  Fortunately computer and the projector costs, at a given performance point, continue to decrease over time.

 

We are currently working on a seven facet operational prototype.  The seven facets will be a slightly modified subset of the 24 facet polyhedra.  This version will have a nearly 180 degree horizontal and about 60 degree vertical field of view.  We expect this to provide a significant sense of immersion.  The hardware cost of this prototype will be about $50,000 based on the component costs outlined above.  We also expect to construct a 180 degree field of view, five facet subset of the surrounding cylindrical surface.  Its cost will be about $35,000.

 

The projected costs for a 12 facet hemispherical surface would be about $85,000. A 24 facet fully surrounding spherical immersive system would cost about $170,000 using today’s prices.  We expect that these systems will become significantly less expense over the next few years as computer and projector costs fall.  Eventually we expect that the rear-projected screens will be replaced by flat panel displays.

 

Initial Use in Human Environment Design

 

The first real application for these prototypes has been in the evaluation of preliminary designs for the Texas A&M College of Architecture’s Architecture Ranch.  The Architecture Ranch is a recently acquired 12 acre off campus site intended to be used by the College to develop innovative facilities, programs, and activities.  An initial step in the development of this site was a design charrette.  Several design teams created a range of development concepts.  A site plan sketch from one of the teams is shown in Figure 7.  This sketch along with aerial photographs and survey data was used as the basis for two three-dimensional virtual site models; one of the as is undeveloped site, the other of the site developed as envisioned in the Figure 7 sketch.  A visualization of the developed site model is shown in Figure 8.  An immersive visualization of the developed site running on one of the three facet prototypes is shown in Figure 9.  Using the immersive system, one can interactively move through either the developed or undeveloped site and experience it from any location and point of view.  The visualization in Figure 8 and the immersive visualization in Figure 9 both use the same software.  The difference is that in Figure 8 the system has been configured for only one computer and one standard rectangular display screen while in Figure 9 the system is configured to use three computational nodes and the three trapezoidal display screens.

 

Additional applications being developed include the design of health care facilities, construction planning visualization and the visualization of inaccessible historic sites including both their past and current configurations.

 

 

Future Directions

 

The development of commodity based lower cost systems may enable the wide spread use of immersive visualization.  However, we see the need for dramatic improvements in ease of use.  Availability of more off-the-shelf applications and workflow integration into a broader range of domains are the real enablers of this technology. Currently, except in a few specialized domains, new uses of these systems requires too much effort in terms of application software development and data preparation.

 

High cost and limited access have inhibited the integration of these systems into routine workflows.  Now that the cost barriers are falling, the development of much better software support and workflow integration is needed for widespread adoption.

 

We expect continued development directed toward more robust, lower cost implementations.    Incorporation of higher performance visual computation, better display technologies, network based collaborative visualization, and developing operational compatibility with other immersive visualization systems and design software is anticipated.

 

In addition to the 12 and 24 facet polyhedra, there are many other polyhedra formed from larger numbers of identical planar shapes.  Two of these, with 60 facets each, are the Deltoidal Hexecontahedron and the Pentagonal Hexecontahedron. These and polyhedra with even larger numbers of facets could be the basis for future modular immersive systems.  In fact, the larger the number of facets, the larger the number of visual computing nodes used.  This increases the overall aggregate resolution and visual computational power of the system.  It also decreases the size of each display facet for an environment of given size.  More facets also result in better approximations to the ideal spherical environment.

 

Faceted display configurations hold the promise of truly modular systems where the immersive environ­ment is created by literally bolting together mass replicated modules.  Each module would contain the required structural, computational, and display elements. The display elements of these modules might eventually utilize flat panel display technology such as organic LEDs.

 

Acknowledgments

 

Our prototype development work is supported by a grant from the Texas A&M University Telecommunications and Informatics Initiative.  Figures 1, 2, and 3 were created by Christina Garcia. The three-dimensional site model used for figures 8 and 9 was developed by Kevin Singleton.  Christopher Anderson and Kevin Singleton developed the software for modular immersive visualization. Christina, Christopher and Kevin are students in the Texas A&M Visualization Sciences graduate program. The sketch shown in Figure 7 is by Marcel Erminy.

 

 

Figure Captions

 

Figure 1  - How a 5 meter diameter 24-facet immersive display structure might be located within a high ceiling room.  The entrance shown is about 6 meters wide.

 

Figure 2 – Projector placement for a 24-facet fully immersive system.

 

Figure 3  - A simulated cross-sectional view of a 5 meter diameter 24 facet immersive display.

 

Figure 4 – A three-quarter scale structural prototype for a 24 facet immersive visualization system.  The diameter of the prototype is about 3.5 meters.

 

Figure 5 – The structure for a three facet operational prototype based on a four facet subset of the 24 facet polyhedra.  Only the top three facets are used for display.

 

Figure 6 – An operational prototype based on a three facet subset of a ten facet surrounding cylindrical surface.

 

Figure 7 – Design charrette site plan sketch.

 

Figure 8 – Three-dimensional environment visualization based on the charrette sketch.

 

Figure 9 – Environment visualized using the three facet immersive prototype system.

 

 

 

References

 

_____, 1987. 3Space Isotrack Users Manual. Polhemus Navigation Sciences, Colchester, VT.

 

Bierbaum, A., et al., 2001. VR Juggler: A virtual platform for virtual reality application development. In: Proc. IEEE VR 2001. Yokohama, Japan.

 

Cruz-Neira, C., D. Sandin, T. DeFanti, 1993. Surround-screen projection-based virtual reality: The design and implementation of the CAVE. In: Proc. Siggraph 93. ACM, New York, pp. 135-142.

 

Dugdale, J., et al., 1999. Current developments in visual display technology for fighter type simulators. In: Proc. 21st IITSE Conf. Orlando, Florida.

 

Hekman, J. and Staff, 1997. Linux in A Nutshell. O'Reilly & Associates, Cambridge, MA.

 

Hereld, M., I. R. Judson, R. L. Stevens, 2000. Introduction to building projection-based tiled display systems. IEEE Computer Graphics and Applications, 20, 4: 22-28.

 

Holden, A., 1971. Shapes, Space and Symmetry. Dover, New York.

 

Humphreys, G., et al., 2001. WireGL: a scalable graphics system for clusters.  In: Proc. Siggraph  2001. ACM Press, New York, pp.129-140.

 

Humphreys, G., et al., 2002. Chromium: A stream processing framework for interactive rendering on clusters. ACM Transactions on Graphics, 21, 3:693-702.

 

Iwata, H., 1996. Rear-projection-based full solid angle display. In: Proc. ICAT ’96 Conf. Makuheri, Chiba, Japan.

 

McCutchen, D., 1991. Method and apparatus for dodecahedral imaging system. U.S. Patent #5,023,725.

 

_____, 2004. http://www.opensg.org

 

Parke, F. I., 2002. Next Generation Immersive Visualization Environments.  In: Proc. SIGraDi 2002 Conference. Caracas, pp. 162-166.

 

Rasker, R., et al., 1999. Multi-projector displays using camera based registration. In: Proc. IEEE Visualization 99, San Francisco, CA, pp. 161-168.

 

Reno, B. A., 1989. Full field of view dome display system. In: Proc. AIAA/FSTC, pp. 390-394.

 

Schaeffer, B. and C. Goudeseune, 2003. Syzygy: Native PC Cluster VR. In: Proc. IEEE VR 2003, Los Angeles, CA,  pp. 15-22.

 

Sterling, T. L., J. Salmon, D. Becker, D. Savarese, 1999. How to Build a Beowulf.  MIT Press, Cambridge, MA.

 

_____, 2002. Viewperf graphics performance benchmark results, <http://www.specbench.org/gpc> (Aug.  2002).

 

_____, 2004. http://www.vrjuggler.org

 

Wenninger, M. J., 1971.  Polyhedron Models. Cambridge University Press, London.

 

Woo, M., J. Neider, T. Davis, 1997. OpenGL Programming Guide, 2nd Ed. Addison Wesley, Reading, MA.