perm filename RADC.NOV[DMA,HE] blob
sn#688222 filedate 1982-11-24 generic text, type C, neo UTF8
COMMENT ā VALID 00002 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 EVALUATION, DETAILED DESIGN, AND IMPLEMENTATION EXAMPLES
C00016 ENDMK
Cā;
EVALUATION, DETAILED DESIGN, AND IMPLEMENTATION EXAMPLES
--------------------------------------------------------
FOR A RULE-BASED
----------------
AUTOMATED STEREO MAPPING SYSTEM
-------------------------------
PROGRESS REPORT
November 19, 1982
Stanford Artificial Intelligence Laboratory
RADC POSTDOC MILESTONES AND DEMONSTRATION
-----------------------------------------
Our last submission to RADC was a timetable of research milestones, leading to a
demonstration of several modes of stereo mapping technique. We report here on
our accomplishments to date.
Review of Contract Research
---------------------------
The focus of the research over this postdoctoral contract has been to augment
and refine the specification of image matching strategies to be employed in our
ultimate stereo mapping system. We do this both with theoretical analyses and
with hand and automated analyses of specific matching strategies applied to both
real and synthesized imagery examples. The demonstration, to be held at the end
of the four month contract (running from August 15 through December 15), was
designed to bring together results from several analysis approaches. As such it
will provide both a demonstration of current manual and automated mapping
abilities, and a chance to assess their applicability to our planned system.
These tests, and associated trial implementations of selected elements of the
stereo mapping system, will provide design exercises leading to detailed design
of major parts of the system. The demonstration outline follows:
Demonstration
-------------
a) the hand synthesis of extended edges,
b) rule-based manual matching of these edges [Binford 81],
c) automated edge extraction [Marimont 82];
rule-based manual matching of edges from an automated
process,
d) automated matching using Baker's system [Baker 82],
e) automated matching using Baker's system with Marimont's
extended edges,
and, if construction work and time permit,
f) computer-assisted stereo correspondence on the stereo station
[Liebes 77].
(Because of building construction underway in the laboratory which
houses the stereo station, the stereo station is not likely to be
available in time for scheduled demos.)
Summary of Progress
-------------------
Edges have been obtained from the automated process of Marimont for the first
set of demonstration data (building complex). Work is proceeding on the second
set (Nasa Ames site), results of which should be available at the end of the
week of Nov 15 (roughly to schedule). The data from this process will be used in
both the manual matching tests and in the automated matching of Baker's system.
Significant alteration was required in the Marimont code in making it provide
image intensity information necessary for processing in the Baker system.
A restructuring of the Baker system for automated stereo mapping has been
finished. Testing has begun on the two data sets. Results from the building
complex should be available at the end of the week of Nov 15, and those from the
Nasa site by the following week (again, to schedule). Complications may arise
with this in that neither set of imagery is in a collinear epipolar frame, and
both may need to be transformed before good matching results can be obtained.
Modifications are being incorporated into this system to allow it to use the
edges derived from Marimont's edge processing. Tests with this data will begin
toward the beginning of the next week.
The analysis of epipolar image registration is continuing. The interest here is
in providing a means for determining corresponding epipolar lines from edge
matchings specified either automatically or manually. Current effort is
directed at obtaining a computationally stable implementation for determining
the epipolar transforms. The difficulty here is in chosing an algorithm that is
relatively insensitive to the common degeneracy of aerial viewing: points used
for camera solving lie at about the same distance from the imaging point. Here,
the discreteness in the sample points' two-space positionings leads to large
uncertainty in the imaging point's three-space positioning. Since the data are
approximately planar, there is not a good depth solution, that is, camera
parameters are not well-determined; but relative depth is determined without a
problem. This may lead to difficulties in numerical solution, not in resultant
accuracy. We are investigating several solution algorithms.
Work has begun on rule-based manual matching of hand synthesized edge data.
Our work with hand synthesized data depends on our use of a program that
operates a digitizing tablet. This program allows interactive construction of
line lists. We had thought the existing program adequate, but on beginning use
of it, found that significant improvement was required to meet our needs. This
has been begun, and should be completed within the next ten days. The principal
improvement is that it should give a line-based network description of image
intensity contours; the previous program did not produce a network description.
We are dependent here on a student working part time. Delay on this puts us
about three weeks behind schedule on the production of hand synthesized edge
data; overlap of other tasks, and intentional slack in the scheduling make this
delay acceptable.
A revised timetable and list of milestones follows:
Milestones
----------
Automated segmentation
Building complex Nov 15
Nasa-Ames Nov 26
Stereo station analysis for ground truth
building complex(conditional) Nov 22
Automated matching with Baker system
building complex Nov 22
Nasa-Ames Nov 29
Implementation of digitizing facility Nov 29
Hand Digitization of edges
Nasa-Ames stereo pair Nov 24
building complex Nov 29
Registration of imagery from terminations
and vertices: Nasa-Ames Nov 29
building complex "
Simulation of rule system for correspondence
Nasa-Ames Dec 6
building complex "
Simulation of rule system for correspondence
using Marimont edges
building complex Dec 10
Nasa-Ames Dec 10
Automated matching with Baker system,
using Marimont edges
building complex Dec 10
Nasa-Ames Dec 10
Examination of OTV inference rules as
ACRONYM rules(possible implementation) Dec 10
Demonstration of matching systems Dec 17
Hand synthesis Automated edge Automated stereo Stereo station Registration
of edge data extraction(Marimont) matching (Baker) analysis
Nov 1 | | | | |
Nov 8 | | | | |
Nov 15 | rule-based X | | |
| matching X | | |
Nov 22 | | ____________/ \_______________ X X |
Nov 29 X | / \ X OTV's in ACRONYM X
Dec 6 | / \ | |
Dec 10 X- -X X
Dec 17 Demonstration
Sincerely
Thomas O. Binford Harlyn Baker
References
----------
[Baker 1982] Baker, H. Harlyn, "Depth from Edge and Intensity Based Stereo,"
Stanford Artificial Intelligence Laboratory, AIM-347, September 1982.
[Binford 1981] Binford, Thomas O., "Inferring Surfaces from Images," Artificial
Intelligence, vol 17(1981), August 1981, 205--244.
[Liebes 1977] Liebes Jr., S. and A.A. Schwartz, "Viking 1975 Mars Lander
Interactive Computerized Video Stereophotogrammetry," Journal of
Geophysics Research, 82, no 28, 4421, Sept. 30, 1977.
[Marimont 1982] Marimont, David H., "Segmentation in ACRONYM," Proceedings
of the ARPA Image Understanding Workshop, Stanford University,
September 1982.