Parallel Computer Vision
Jon A. Webb
Table of Contents
2. Current Work
3. Future Work
This project applies advanced, low-latency supercomputers to problems in computer vision. Click here for a photographic guide to the people in the lab (with huge images; not for the bandwidth-limited.)
Early work led to the use of the Carnegie Mellon Warp machine in the Navlab robot vehicle. A Warp machine was mounted in Navlab and used for various tasks, including road following using color-based image segmentation, and also using the ALVINN neural-network system.
The ALVINN work was particularly successful, because an important factor in developing the road following neural network was the availability of Warp to perform back propagation on large neural networks.
More recent work has been centered around the iWarp computer developed jointly with Intel Corporation. An FTP directory of iWarp papers and iWarp home page are available.
- Managing large data structures. We (George Gusciora, Webb, and H. T. Kung) are studying how algorithms that manipulate large data structures can be mapped efficiently onto a distributed memory parallel computer, in a Ph.D. thesis expected in January 1994.
- Efficiently moving data structures. We (Doug Smith, Webb, and H. T. Kung) are studying how can the most efficient of the many possible communications patterns for reorganizing data structures within a parallel computer be generated automatically, in a Ph.D. thesis expected in 1994.
- Developing parallel vision software. We are implementing a parallel version of the ISO standard Programmer's Imaging Kernel System. This implementation is based on Adapt, which has been mapped onto the Carnegie Mellon-Intel Corporation iWarp computer and the Intel Paragon (by Mike MacPherson of Intel Corporation). An implementation for the IBM SP1 is planned (and will be done by George Gusciora at the Maui High Performance Computing Center.)
- Real-time stereo vision. We (Webb and others in the iWarp group at Carnegie Mellon University, including Tom Warfel, who built the video interface, and Dave O'Hallaron) have implemented the fastest stereo vision system ever demonstrated. It uses Kanade-Okutomi multi-baseline stereo and operates at 15 Hz on a 64-cell iWarp, turning three 240x256 input images into a 240x256x16 depth image. This system was recently demonstrated at Supercomputing '93.
- Advancing general-purpose programming models. The essential Adapt looping construct has been implemented as a parallel DO loop called PDO in a Fortran variant called FX, which is being developed by a research group at Carnegie Mellon headed by Thomas Gross, Dave O'Hallaron, and Jaspal Subhlok. Student members include James Stichnoth, Bwolen Yang, Peter Dinda, Tom Stricker, Ali-Resa Adl-Tabatabai, Eka Ginting, and Susan Hinrichs. Staff include Petere Lieu and Guei-Yuan Lueh. An FTP directory of FX papers and FX home page are available.
- We (Sing Bing Kang, Webb, C. Lawrence Zitnick, and Takeo Kanade)
have built an parallel active stereo vision system designed to recover
depth images with less than 1 mm error. We are seeking scientific
applications of such a system, which is capable of capturing as
many as several hundred images at 30 Hz for later analysis to recover
depth with less than 1 mm accuracy over a range of meters. This system is
implemented in FX, using in part code developed by Peter Dinda under
the direction of Dave
O'Hallaron. A version
of Dinda's code is available by anonymous FTP. A technical report
that describes the system is available here
Our current interest is in building computer vision systems that use parallel computers to achieve performance never previously demonstrated:
- We (Mark Wheeler, Katsushi Ikeuchi, and Webb) are studying issues in higher level parallel vision in the context of an automatic tool for generating parallel object recognition programs, based on the Vision Algorithm Compiler.
- We (Paul Heckbert, Webb) are working to combine the stereo vision system with high-quality graphical output in the Real, Really Real Graphics project.
- We (Prem Janardhan, Jonathan Minden, Webb, and Katsushi Ikeuchi) are applying parallel image processing and computer vision techniques to problems in biology as part of the Automated Interactive Microscope project (a large project that is headed by Scott Fahlman and Lans Taylor of Carnegie Mellon University.)
- Finally, click here for a large (432 KB)
Pro-cite refer format bibliography of papers in parallel computer
vision and related areas.