Despite significant progress in automatic recovery of static scene structure from range images, little effort has been made toward extending these approaches to dynamic scenes. This disparity is in large part due to the lack of range sensors with the high sampling rates needed to accurately capture dynamic scenes. We have developed a system that overcomes this problem by exploiting video cameras, which easily capture images of dynamic scenes, and image-based stereo, which estimates scene structure based on image correspondences. Our system uses a synchronized multi-camera recording system to capture live video of the scene and an off-line image-based stereo process to compute range. By combining this system with multi-image fusion, we created a novel system for dynamic structure recovery. This output is used in the Virtualized Reality project, which immerses a viewer in images synthesized from a viewer-controlled virtual camera.
Back to the workshop.
ModelMaker is a hand-held sensor for rapid capture of realistic colour 3D object models. The ModelMaker 3D scanning and colour acquisition system were developed by 3D Scanners Ltd. Geometric fusion software for integrating 3D surface measurements into a single object model was developed at the University of Surrey.
ModelMaker uses a 6DOF positioning system such as a Faro arm or a Pixsys tracker to locate the laser stripe sensor in space. As you scan, a rendered image appears on the monitor to show you what has been covered. The process is like painting.
Fusion of the scanned surface measurements into a single 3D model is based on an intermediate volumetric implicit surface representation. Local geometric constraints based on measurement uncertainty ensures reliable surface reconstruction. Integration uses operations in 3D space allowing fusion of sets of surface measurements for arbitrary objects. This avoids limitations on surface geometry inherent in previous techniques based on a 2D range image structure or 2D projection of the data. Implicit surface based geometric fusion enables fast and accurate reconstruction of complex objects from either conventional range images or a hand-held range sensor.
Back to the workshop.
We present an algorithm that can efficiently create closed triangular meshes from 3-D geometric data. This data can be presented in the form of images or unordered points. The mesh model can be created to any desired accuracy relative to the original data points. The created mesh can also be closed to make a true volumetric model.
Back to the workshop.
In this talk I will present our new approach for modeling and rendering architectural scenes from a sparse set of still photographs. The modeling approach, which leverages both geometric and image-based techniques, has two components. The first component is a photogrammetric modeling method which facilitates the recovery of the basic geometry of the photographed scene. Our photogrammetric modeling approach is effective, convenient, and robust because it takes advantage of the constraints that are characteristic of architectural scenes. The second component is model-based stereo, which recovers how the real scene deviates from the basic model. By making use of the model, this stereo technique robustly recovers accurate depth from widely-spaced image pairs. Consequently, our approach can model large architectural environments with far fewer photographs than current image-based modeling approaches. For producing renderings, I will present view-dependent texture mapping, a method of compositing multiple views of a scene that better simulates geometric detail and specular reflectance than flat texture-mapping. I will present results that demonstrate our approach's ability to create realistic renderings of architectural scenes from viewpoints far from the original photographs, including images from the Rouen Revisited art installation presented at the SIGGRAPH '96 art show.
For more info, see: Modeling and Rendering Architecture from Photographs.
Back to the workshop.
Key to any automated surface acquisition system is an ability to reposition a range scanner so that its next range image will sample previously unseen portions of the viewing volume, that is, the system must be able to solve the "next best view" (NBV) problem. From its next position, the scanner should not only have the possibility of sampling more of the object's surface but should also resample part of the object already scanned to allow for the registration and integration of the new range data with the previous scans. A novel representation, positional space, is presented which facilitates a solution to the NBV problem by representing what must be and what can be scanned in a unified data structure. The expensive operations involved in determining the visibility of any part of the viewing volume are computed only once, not for each potential position of the scanner, thus breaking the computational burden of choosing the NBV from a large number of positions. No assumptions are made about the geometry or topology of the object. The algorithm will scan all visible surfaces of an object, can be directed to resample surfaces which were scanned with low confidence, and is (theoretically) self terminating. In addition, the algorithm will work with nearly any dense sampling range camera and scanning setup. A completely automated surface acquisition system featuring the proposed NBV algorithm is presented and demonstrated on a real object.
Back to the workshop.
Scalar (and vector) fields arise in several scientific applications. Existing scalar field reconstruction and visualization techniques require that the user infer the global scalar structure from what is frequently an insufficient sampling or display of information. We present a reconstruction and visualization technique which numerically computes the topological structure at all scales, removing from the user the responsibility of extracting information implicit in scanned field data, and presenting the structure explicitly for analysis. We further demonstrate how scalar topology detection proves useful for correct image processing and visualization applications.
Back to the workshop.
Garments of a particular size are typically assigned to inductees by an unscientific "you look like this size" approach. Currently available technology provides for three-dimensional, non-contact measurement of the body with a high level of accuracy. With the appropriate software, dimensions extracted from this measuring system could be used to sort through available sizes to choose a best fit. In addition, these dimensions could be used to adapt existing patterns specifically for an individual inductee. These patterns could be input into a CAD system whereby individual markers, single-ply cutting, and modular manufacturing could produce individually-sized garments. The objective of this effort is to determine whether the application of three-dimensional, non-contact measurements can improve the fit of military uniforms and prevent the costs of alterations. These improvements would be achieved through providing a better fit when the uniforms are initially assigned and would encourage the use of contemporary apparel manufacturing techniques.
Back to the workshop.
The development of laser-based surface digitizers promises to provide the engineer and clothing designer with abundant and unique data on the human form. Anthropologists at the U.S Army Natick Research, Development, and Engineering Center are exploring the use of whole body and head 3D digitizers in development of personal protective equipment and military clothing. We will report on current applications which include estimation of body armor surface area coverage, and extraction of standard anthropometric measurements from 3D scans.
Back to the workshop.