RADIUS BAA Research -- SRI International

The construction and use of 3D models of military and industrial sites will allow revolutionary advances in the speed, confidence, and range of analytical techniques with which an Image Analyst develops and reports intelligence information. This project seeks to increase the speed and accuracy with which site models can be constructed from current imagery by developing a new family of image understanding (IU) techniques. Our research is proceeding on two fronts simultaneously:

Model-Based Optimization

Model-Based Optimization (MBO) algorithms, in which an objective function is optimized to determine the best fit with image data, provide an ideal basis for semiautomated site modeling. The so-called "snake" technology (first popularized by Witkin, Terzopoulos, and Kass) is a particular example of model-based optimization that is named for the way the curves wiggle during optimization. SRI has been a pioneer in its development, having designed and implemented numerous techniques for finding such objects as roads, buildings, vegetation regions, and rivers. The ability to fit a deformable object model to multiple images simultaneously, and the employment of continuation methods to avoid local minima, are examples of our advances.

SRI is extending the MBO technology to provide the capability of extracting many different object classes under a wide variety of imaging and scene conditions. When integrated into the RCDE, this technology will constitute a suite of feature extraction tools tailored to the needs of the intelligence analyst.

[image] (full size image 285k)

The SRI MBO system extracting a 3D model of a building from two images.

[image] (full size image 315k)

The SRI MBO system extracting a 3D model of a road from an overhead image. The elevation at each point of the road is calculated by intersecting the camera ray with the digital terrain model (DTM).

Further details can be found in:

  • Parametric Models are Versatile: The Case of Model Based Optimization.

    Context-Based Vision

    We are investigating the design of an architecture that can be used as the basis to control the invocation of IU algorithms for feature extraction. The key research question is whether sufficient contextual constraints are available to choose the algorithms and parameters that are necessary for site model construction.

    Our approach has been to apply the context-based architecture incorporated in CONDOR, an SRI system for automatically constructing scene models of natural terrain from ground-level views. The semiautomated nature of RADIUS allows access to additional sources of contextual constraints that were not available to CONDOR. We are adapting the CONDOR mechanisms to better suit the interactive nature of RADIUS. A technical paper on the state of that research was published in the Proceedings of the ARPA IU Workshop held in April, 1993.

    Further details can be found in:

  • Integrating IU Algorithms into the RADIUS HUB

    SRI Quarterly Reports:

    Aaron Heller <heller@ai.sri.com>
    Last update: Thurs Nov 9 1995
    Copyright © 1994, 1995 SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025 USA. All rights reserved.