OCT.1
SCAAT: Incremental Tracking with Incomplete Information
Greg Welch

Abstract I will present a new Kalman filter-based method for tracking a user's position and orientation in virtual environments. The method uses a sequence of single sensor observations (constraints) as opposed to groupsof observations. I refer to this new approach as single-constraint-at-a-time (SCAAT) tracking.

As compared to current multiple-constraint approaches, the SCAAT method improves tracking accuracy, timing, and flexibility. It improves accuracy by properly assimilating sequential device observations, by filtering sensor measurements, and by autocalibrating system parameters concurrently while tracking. It improves estimate rates and latencies by producing a new estimate with each new individual device observation. It offers flexibility by facilitating user motion prediction and multisensor data fusion.

We are applying the SCAAT approach to 3D tracking for virtual environments. However I believe that this work may prove to be of interest to the larger scientific and engineering community in addressing a more general class of tracking and estimation problems.

Cornell University Psychophysics is the quantitative study of the relationship between the physical properties of the world and their perceptual appearances. Physical objects and events have measurable properties like area, weight, and intensity, and when we perceive these objects we say they have a certain size, or are so heavy, or so bright. But while we can directly measure the physical properties, we can only infer their appearances from observers' subjective responses. So psychophysics offers a set of unbiased experimental procedures that allow researchers to indirectly measure the percepts caused by different physical stimuli. The results of psychophysical experiments can be used to establish predictive mathematical models of perception that can be used in graphics, image processing, and computer vision. I will give a broad introduction to the field of psychophysics, starting with classical methods for measuring perceptual thresholds, then describing advances in signal detection theory, and ending with a discussion of psychophysical and psychometric scaling methods.

This talk will present ongoing research in gestural user interfaces for 3D modeling. Included will be a discussion of the Sketch system for rapidly creating approximate 3D models. Also covered will be up-to-date research in two-handed interfaces, as well as non-realistic rendering.

A number of techniques have been proposed for flying through scenes by redisplaying previously rendered or digitized views. Apple's QuickTime VR is one example. In this talk, I will describe a simple and robust method for generating new views from arbitrary camera positions without depth information or feature matching, simply by combining and resampling the available images.

This is joint work with Pat Hanrahan. For those who saw Pat's Siggraph '96 presentation, this talk will be a longer version. In particular, I will focus in more depth on the following issues:

• o Alternative ways of parameterizing light fields
• o Line space interpretations, ray coverage, and sampling uniformity
• o Technical challenges related to the design of camera gantries
• o Implementation and performance of our VQ compression method

Finally, I will summarize our plans to construct two practical "light field cameras", one that employs a single camera and a 4-DOF motion system, and another based on a planar array of CCDs with compression chips interspersed.

DEC.3
DIS & telecollaboration
Mike Macedonia

We're sorry...
No abstract is currently available.

DEC. 5
Information Visualization and the WWW
James Foley, MERL

Information Visualization involves displaying data and data relationships that are primarily non-numeric and non-geometric, in contrast to its cousin Scientific Data Visualization, in which most or all of the data is numeric and already has geometric position.

Information Visualization can be used with the World-Wide Web and similar information spaces in two ways: in navigating through the information space to understand how different pieces of information relate one to another, and in presenting the results of queries returned by search engines.

But, navigating the Web (cruising the infobahn) currently bears almost no resemblance to navigating the real world. The Web is primarily a linguistic world; the real world is visual. When I drive from city to city, I have a visual tool, the road map, to help me find my way, aided by linguistic aids (road signs) along the way. We describe research progress toward automatically creating visual road maps for the WWW.

Similarly, search results are generally presented as a textual list, ordered by relevance. The ordering does little to help the user understand why each document has been retrieved. We describe a visualization technique that may help users understand document relevance.

In closing, we discuss potential enhancements to the WWW infrastructure which would increase its ability to create and present visualizations.

Jan 23
Visualization of Multi-valued Volume Data
David Laidlaw

Multi-valued volume data has become more prevelant as data acquisition and computers have become faster and as limitations of scalar-valued measurements are reached. I'll begin with a video of some recently acquired multi-valued data from biology, medicine, and fluid mechanics, and talk about three specific problems that we've been addressing in that context.

The first problem is understanding developing mouse embryos through MR imaging. The MR images produced are vector valued, and extracting maximal information from the data is difficult. I'll present a goal-based technique for choosing linear combinations of the scalar elements making up the vector-valued data. The goals that we have implemented look for combinations that are useful for volume rendering; the technique can be easily extended to other goals.

The second problem is tissue classification of vector-valued data. This problem is a refinement of the first, and I will describe our algorithm, which is geared towards making geometric models of biological samples from MRI measurements.

The third problem is understanding the progression of a mouse analogue of multiple sclerosis. I will describe a new MR imaging protocol that we have been developing that measures the diffusion tensor field over a sample. This second-order tensor field shows significant geometric structure, some of which correlates well with histological sections of mice spinal cords, and the technique is much less invasive than sectioning. I'll try to give some geometric intuition about second-order diffusion tensors, and touch on the display of the data. The new imaging protocol also has more general applications in understanding musculature and central nervous system structures.

Jan 30
Inverse Image Warping
Gary Bishop

I will describe the preliminary work Leonard McMillan and I have been doing on an inverse implementation of McMillan's imaging warping equation. Previous implementations have been "forward" in that they have transformed pixels from the reference (input) image to the desired (output) image and then attempted to solve the reconstruction problem in the output image. Our inverse approach, like standard texture mapping, transforms the coordinates of output image pixels into the reference image and handles the sampling issues there. The inverse approach may allow us to think more clearly about the sampling and reconstruction issues in image-based rendering. It also makes it simple to eliminate some artifacts of image-based rendering by merging multiple reference images to form a single output image.

I will demonstrate an interactive JAVA based inverse warper.

February 18
The Topology of Implicit Surfaces
John C. Hart

Implicit surfaces offer a powerful geometric representation that simplifies the smooth blending of arbitrary shapes, but they also have historically been difficult to model interactively, render efficiently, and incorporate into production graphics systems. Converting an implicit surface into a polygonal mesh overcomes many of these problems, but polygonization sometimes does not accurately represent the structure of the implicit surface.

Treating an implicit surface as a gradient system allows theorems from catastrophe theory and Morse theory to accurately determine its topology by examining the critical values of the system. Techniques based on interval analysis find the critical points of the system, and dynamic constraints can be used to track these critical points as the implicit surface changes. Techniques for modifying the polygonization to accommodate changes in the implicit surface topology are given. These techniques are robust enough to guarantee the topology of an implicit surface polygonization, and are efficient enough to maintain this guarantee during interactive modeling. The impact of this work is a topologically-guaranteed polygonization technique, and the ability to directly and accurately manipulate polygonized implicit surfaces in real time.

This research is the topic of the Ph.D. dissertation of Bart Stander, a WSU grad student currently employed at Strata.

February 20
Simplifying Polygonal Models
Jon Cohen, Dinesh Manocha

UNC's Walkthrough research group has recently focused on fast ways to render complex environments, such as a submarine or tanker vessel. We shall discuss what role geometric simplification plays in the rendering of such scenes and also mention some other important techniques.

This talk will describe some current methods for simplifying polygonal models. It will focus on the Simplification Envelopes technique, which provides strict error bounds on the deviation of the simplified surface from the original surface. Such bounds allow the automatic selection of switching distances that determine when it is appropriate to view a simplified model.

The talk will also venture into the realm of some more recent work. This work deals with computing successive mappings as the surface is simplified. It uses techniques from computational geometric, such as linear programming, to solve some of the sub-problems that can arise for many simplification techniques. We are hopeful that such techniques may help us provide strict error bounds in other domains (e.g. color, normal, texture) as well.

Suggested reading:

Cohen, Jonathan, Amitabh Varshney, Dinesh Manocha, Greg Turk, Hans Weber, Pankaj Agarwal, Frederick Brooks, William Wright. "Simplification Envelopes". Proceedings of SIGGRAPH 96(New Orleans, LA, August 4-9, 1996). In Computer Graphics Proceedings, Annual Conference Series, 1996, ACM SIGGRAPH, pp. 119-128.

ftp://ftp.cs.unc.edu/pub/users/cohenj/Simplification/SimpEnv.SIG96.8bit_1 00dpi_i mages.ps.gz

Other usefule sources of information:

February 25
Trends in PC Graphics
Turner Whitted

This talk describes the evolution of 3D graphics technology as it migrates to personal computers and describes the changes to the personal computer as it absorbs 3D graphics. The discussion is limited to personal computers which use the Intel CPU family and concentrates on hardware rather than the crazy quilt world of competing software standards. However, I intend to tackle at least one or two controversial questions about both hardware and software.

March 6
Hybrid Tracking
Greg Welch, Gary Bishop, Andrei State

Position and orientation tracking for virtual environments can be accomplished using a variety of mediums or modalities, e.g. electromagnetic waves, sound, or radio waves. However, when used alone each of these modalities suffers from inherent drawbacks. To address these drawbacks some researchers have sought to develop hybrid systems that leverage the strengths of distinct modalities to maintain more consistent performance throughout a working environment, across the frequency spectrum, and over a wide range of dynamics.

During this talk Welch and Bishop will explore the strengths and weaknesses of several common tracking modalities, present and discuss some past and present hybrid systems, and offer a glimpse into the direction of work under a newly funded joint effort between UNC, USC, and Hughes Research Labs. Andrei State will complete the talk with an overview of the UNC magnetic-optical hybrid system developed by our Ultrasound group.

April 3
Mark Mine
Working in a Virtual World

In this talk I will present ongoing research in immersive virtual- environment (IVE) interaction techniques. Included will be a discussion of several issues of working in a virtual world, examples of effective IVE interaction techniques and an overview of some principles of working in a virtual world. Examples will be given from the Chapel Hill Immersive Modeling Program, CHIMP.
For more information visit Mark Mine's home page:

April 8
Greg Welch, Gary Bishop
The Kalman Filter

The Gaussian concept of estimation by least squares, originally stimulated by astronomical studies, has provided the basis for a number of estimation theories and techniques during the ensuing 170 years-probably none as useful in terms of today's requirements as the Kalman filter.

H. W. Sorenson
University of California, San Diego
IEEE Spectrum, vol. 7, July 1970

In 1960, R.E. Kalman published his famous paper describing a recursive solution to the discrete-data linear filtering problem [Kalman60]. Since that time, due in large part to advances in digital computing, the Kalman filter has been the subject of extensive research and application, particularly in the area of autonomous or assisted navigation.

The Kalman filter is a set of mathematical equations that provides an efficient computational (recursive) solution of the least-squares method. The filter is very powerful in several aspects: it supports estimations of past, present, and even future states, and it can do so even when the precise nature of the modeled system is unknown.

For this STC lecture our goal is to provide a practical one-hour introduction to the Kalman filter. This includes the following:

• some intuition about the filter operation,
• the introduction of the basic discrete Kalman filter equations,
• a relatively simple (tangible) example with real numbers & results,
• highlights of some applications, and
• a Kalman filter handout and pointers to more detailed reading.

A very "friendly" introduction to the general idea of the Kalman filter can be found in Chapter 1 of [Maybeck79], while a more complete introductory discussion can be found in [Sorenson70], which also contains some interesting historical narrative. More extensive references include [Gelb74], [Maybeck79], [Lewis86], [Brown92], and [Jacobs93].

References

Kalman60:

Kalman, R. E. 1960. "A New Approach to Linear Filtering and Prediction Problems," Transaction of the ASME-Journal of Basic Engineering, pp. 35-45 (March 1960).

Maybeck79:

Maybeck, Peter S. 1979. Stochastic Models, Estimation, and Control, Volume 1, Academic Press, Inc.

Sorenson70:

Sorenson, H. W. 1970. "Least-Squares estimation: from Gauss to Kalman," IEEE Spectrum, vol.7, pp. 63-68, July 1970.

Gelb74:

Gelb, A. 1974. Applied Optimal Estimation, MIT Press, Cambridge, MA.

Lewis86:

Lewis, Richard. 1986. Optimal Estimation with an Introduction to Stochastic Control Theory, John Wiley & Sons, Inc.

Brown92:

Brown, R. G. and P. Y. C. Hwang. 1992. Introduction to Random Signals and Applied Kalman Filtering, Second Edition, John Wiley & Sons, Inc.

April 10
Elaine Cohen and Rich Rienenfeld
Bases, Representations, and Design Methods in Mathematical Modeling

Abstract to Be Announced

April 15
Andries van Dam and Sascha Becker
Virtual Reality Modeling Language in Context

The purpose of this talk is to introduce the history, technology, and applications of VRML, the Virtual Reality Modeling Language. VRML arose as an Internet standard for 3D graphics in 1995; over the last two years it has spawned a fledging industry, complete with a proposed ISO standard and a consortium to promote it. VMRL 2.0, the current version, is a file format for 3D objects and scenes with behaviors. A VMRL scene is described by a hierarchical scene graph, and a simple event model drives behaviors in the scene. Simple interactions are supported directly in VMRL, and more complex interactions or behaviors may be built with Java or JavaScript. We will compare VRML to several modern graphics APIs, including OpenGL and Java3D. We will then survey some of the tools available for browsing and authoring VRML worlds or developing VRML applications. Finally we will present resources for more information on VRML.

April 17
Density Estimation Techniques for Global Illumination
Bruce Walter

This talk will describe the density estimation framework developed at Cornell to produce view-independent global illumination solutions that are suitable for interactive walkthroughs. The method consists of three pieces: a particle tracing phase probabilistically simulates the flow of light, a density estimation phase reconstructs the lighting function from the particle data, and a decimation phase compacts the results for faster display and reduced storage.

The first part of the talk will describe some of the advantages of the framework including: scalability, natural parallelism, and robustness, and some of its current drawbacks. The second part will concentrate on current and future research on the density estimation phase. Density estimation is a standard statistical problem and some results from the statistics literature will be discussed along with ideas for extending it to handle wider classes of models and for extracting more information from the particle data.

Publications:

Peter Shirley, Bretton Wade, David Zareski, Philip Hubbard, Bruce Walter and Donald Greenberg,

"Global illumination via density estimation", In Proceedings of the Sixth Eurographics Workshop on Rendering, June 1995

Bruce Walter, Philip Hubbard, Peter Shirley, and Donald Greenberg,

"Global illumination using local linear density estimation", In ACM Transaction on Graphics, To appear summer 1997