Continuing Obstacles to Acceptance of Optical Coordinate Measurement
Everyone in the field knows the wondrous technical advantages of 3D optical coordinate measurement. Optical scanners can be 10 million times faster than conventional methods of measuring 3D points. If there existed commercial enterprises that were completely point-acquisition-limited (PAL), it might seem logical that they could increase their throughput (and revenues) by a factor of 10,000,000 with the right combination of technology, most of which really exists today. So why haven't 3D optical scanner vendors sold thousands or millions of scanner units based on the huge productivity advantages? Why are 3D scanners nowhere near as common as 2D and 1D scanners? While no one has all the answers and some may argue with some of the premises, this talk will briefly review the optical coordinate measurement field since 1977 and discuss some of the issues encountered during that time and list some of the obstacles still remaining.
Back to the workshop
Digitizing and Modeling Animatable Objects
Much of the work on fitting surfaces to digitized data has focused on the problem of accurately reproducing static shape. Animation, however, additionally requires that the model move appropriately and predictably. In this talk I'll describe some work in progress to address various issues that arise when constructing animatable objects from physical models.
Back to the workshop
Surface and Volume Reconstruction
This talk considers the general and vague algorithmic problem of reconstructing a surface from a given finite set of points in space. There are many versions of this problem discussed in the literature differing in their assumptions on the input and the type of generated surface. A valuable bonus is the automatic triangulation of the part of space bounded by the surface.
Standard topological terminology is used to differentiate between surface types and to describe normal and abnormal cases. Emphasis is placed on Delaunay complex based methods for surface and volume reconstruction. Software illustrating Alpha shapes and Wrap surfaces is presented.
Back to the workshop
Automatic reconstruction of B-spline surfaces of arbitrary topological type.
Creating freeform surfaces is a challenging task even with advanced geometric modeling systems. Laser range scanners offer a promising alternative for model acquisition-the 3D scanning of existing objects or clay maquettes. The problem of converting the dense point sets produced by laser scanners into useful geometric models is referred to as surface reconstruction.
In this paper, we present a procedure for reconstructing a tensor product B-spline surface from a set of scanned 3D points. Unlike previous work which considers primarily the problem of fitting a single B-spline patch, our goal is to directly reconstruct a surface of arbitrary topological type. We must therefore define the surface as a network of B-spline patches. A key ingredient in our solution is a scheme for automatically constructing both a network of patches and a parametrization of the data points over these patches. In addition, we define the B-spline surface using a surface spline construction, and demonstrate that such an approach leads to an efficient procedure for fitting the surface while maintaining tangent plane continuity. We explore adaptive refinement of the patch network in order to satisfy user-specified error tolerances, and demonstrate our method on both synthetic and real data.
Back to the workshop
Generating Virtual Worlds from Real Worlds using Computer Vision
Virtual reality simulation (VRS) is a convenient tools for various applications: computer games, military simulation, and robot programming. Currently, most models for these systems have been manually constructed by programmers.
We are developing techniques to construct solid models from observation of real objects and scenes. This technique can bypass expensive programming efforts and can drastically reduce the cost of constructing such models. Since this technique maintains 3D solid models instead of a sequence of images (as in image mosaic systems), we can reduce the amount of memory and render the objects/scenes more accurately.
In this talk, I will present three key technologies recently developed in our group:
Finally, I will overview on-going projects including building a virtual Carnegie Mellon Wean Hall corridor using a movable range finder cart.SAI (spherical attribute image)
PCAMD (principal component analysis with missing data) for obtaining the statistically optimal representation from a sequence of registered (yet noisy) range images, and
PHOTOMETRIC SAMPLING for recovering reflectance characteristics of objects from a sequence of color images.
Back to the workshop
Building computer models models from multiple range images
In this talk I will summarize Stanford's efforts to build a practical system for digitizing the shape and appearance of physical objects. First, I will review our recently published papers in this area, including our volumetric method for combining multiple range images to generate dense watertight polygon meshes (work with Brian Curless), our algorithms for converting these meshes into tensor product B-spline surface patches with accompanying displacement maps (work with Venkat Krishnamurthy), and our methods for analyzing laser reflection image sequences that corrects several well-known problems inherent to laser-stripe scanners (work with Brian Curless). Next, I will outline several ongoing projects: automatically determining the next best view to acquire, acquiring spatially varying color and reflectance, and acquiring large-scale objects using time-of-flight scanners, Finally, I will speculate on the difficulties of building, managing, and displaying complex virtual environments and on the apparent tension that is arising between model-based and image-based graphics.
Back to the workshop
Digitizing Objects With XRAY Computed Tomography
Most of us are familiar with XRAY Computed Tomography as a medical imaging modality that allows radiologists to view internal anatomy. The technology can also be used in non-destructive testing of industrial parts. Scanning parts for CAD is an unexplored application of CT. The practical use of CT for digitizing has been hampered by its cost and availability. This talk will describe the state of the art of CT for medical, quality and digitizing applications. Strengths and weaknesses of CT scanning will be discussed.
Among the topics to be covered are:
Back to the workshop
Building Parts You Could Not Build Before or Everything You Ever Wanted to Know About SDM
This talk will describe a new manufacturing process called Shape Deposition Manufacturing* (SDM). In this process the benefits of material deposition methods such as plasma and laser welding are combined with multi axis numerical cutting. Novel artifacts can be built with the help of SDM. Examples include: Multi-material layers, structures of arbitrary geometric complexity, parts with controlled microstructures, and devices with embedded electronic components and sensors in conformable structures. Important issues toward the production of high quality objects are concerned with the creation of interlayer metallurgical bonding through substrate remelting, the control of the cooling rates of both the substrate and the deposition material, and the minimization of residual thermal stress effects.
Back to the workshop
Recent Developments in 3D Laser Scanning
A presentation of the current research activities in 3D digitizing will be made. Emphasis will be on the digitizing of environment for applications such as documentation and simulation. We will show experimental results related to industrial, museum and space interests. Fundamentals of laser properties for 3D digitizing will also be discussed, in the context of future developments.
Back to the workshop
Next-generation CAD tools
Capturing geometry with a scanning device is only one of several steps in reality capture. Purely geometric information has limited value without added structuring and/or semantic information. Integrating such additional information into a geometry model will require a human user in the model acquisition loop. Thus the system takes on the nature of an interactive CAD tool, a domain that is fraught with opportunities and pitfalls. I will discuss some of the desirable model extensions for various application domains and ways of incorporating such information. I will review lessons learned about abstractions and user interfaces from the domain of integrated circuit layout, as designers stopped pushing around low level mask layout geometry, and started to manipulate ever higher-level primitives.
Back to the workshop
Image Denoising, Edge Detection, and Shape Recovery
Joint work with R. Malladi.
This work presents a unified approach to image denoising, edge detection and shape recovery, based on level set methods and fast marching methods for solving static Hamilton-Jacobi equations. The results are developed in the context of medical imaging and segmentation of tumors, three-dimensional time-depedent reconstruction of MR/CT cardiac data, and an arterial/cortical structures.
Beginning with an image, the first stage of this approach removes noise and enchances the image by evolving the image under flow controlled by min/max curvature flow and by the mean curvature. The noise removal/enhancement schemes applied in this stage contains only one enhancement parameter, which in most cases is automatically chosen, and stop automatically at some optimal point. Continued application of the scheme produces no further change.
The second stage of our approach is the shape recovery of a desired object; we exploit a combination of level set methods and fast marching methods to evolve an initial curve/surface towards the desired boundary, driven by an image-dependent speed function which automatically stops at the desired boundary.
We will show a collection of simulations and movies of denoising/edge detection/shape recovery simulations.
Back to the workshop
Recovering Geometric and Photometric Models from Multiple Images
This talk will overview various techniques we have developed for constructing 3-D models from multiple intensity images (also known as "motion stereo") and discuss their applicability:
Back to the workshop
New 3D Reconstruction Algorithms for Pose-Instrumented Cameras
We are developing a system whose goal is completely automated reconstruction of 3D CAD models of urban exteriors. The input to the system is comprised of "pose-images"; high-quality (2x3K) digital still images, each annotated with accurate estimates of 6-DOF camera pose in an absolute (e.g., Earth) coordinate system. The output is a 3D CAD model of building exteriors, along with a BRDF estimate for each output polygon. Our system has no "human in the loop," except in the input phase in which instrumented images are acquired.
We describe several algorithms which become feasible only when accurate estimates of absolute camera pose are available. These algorithms should make possible automated reconstruction to sub-meter resolution of CAD models representing entire cities, from thousands of high-resolution instrumented images.
Back to the workshop
Measuring Reflectance -- Even If You Can Pronounce Goniospectroradiometry -- Can You Do It?
In this talk, Greg Ward of the Lawrence Berkeley National Lab will discuss his research and experience in capturing reflectance distribution functions and spectra of common materials. He will present several alternative methods for measuring BRDFs, discuss their strengths and weaknesses, and suggest possible alternatives for the future.
Back to the workshop
The Advent of Shape Photography (TM)
Up until now three-dimensional data capture has been limited to methods that capture a full image but are slow compared to photography, such as laser scanning and structured light, and methods that are instantaneous, but capture only isolated points. Shape photography will change all this. A shape camera (TM) instantaneously captures colored three-dimensional images comparable in accuracy and coverage to laser scanning and structured light, in a compact, portable, and relatively inexpensive system. The theory underlying shape photography will be explained, and a shape camera will be demonstrated.
This talk is based on research done with C. Lawrence Zitnick (Robotics Institute, Carnegie Mellon University.)
Shape camera, shape photograph, and shape photography are trademarks of Visual Interface, Inc.
Back to the workshop