

		  Fort Hood Image Distribution
			 Second Data Set
			  Version 2.0
       	                 June 27, 1995


	                 Chris McGlone
                   Digital Mapping Laboratory
	           School of Computer Science
	           Carnegie Mellon University
	             Pittsburgh, PA.  15213	



1. CONTENTS

This tar file contains the second distribution of the second set of 
image data taken of Ft. Hood, TX, for the RADIUS program.  The first
set of Fort Hood imagery was taken on December 10, 1991 and consisted
of four images, two vertical and two oblique.  The second
data set  includes 7 vertical and 19 oblique images,
which cover a larger portion of the Fort Hood base area
and span a far greater range of viewing angles and spatial resolutions.
The images and initial orientations were distributed at the RADIUS 
meeting at SRI, March, 1995.

This distribution contains final orientation information for the
images (the orientation data with the last distribution is now
obsolete).  Due to problems discovered after receipt of the initial
triangulation data, the contractor (PSI) was required to re-do the
resections on the oblique imagery.  Unfortunately, this was also
unsatisfactory. The orientation data in this distribution is from
a simultaneous adjustment of all the images of the
motor pool area, performed at the Digital Mapping Laboratory.
(A more complete discussion of the resection problems is in
section 2.5).

The contents of this tape are:

	README2 (this file)

	image-char-2.{tex, ps}--image characteristics table file containing a
		list of the images and an overview of image information (size, 
		tilt, kappa, GSD, etc).

	all.resect, all-out.resect		
		input and output .resect (control point image measurement)
		files for the block adjustment (see format description below)
	
	all.out, output of the block adjustment program

	for each image:
		TEC image header file, with RADIUS extensions
		FBIP file in ASCII format

1.5 Contents of image characteristics table file

	The image characteristics table contains the image id, the
general area of Ft Hood covered by the frame, and the date it was
taken.  The approximate tilt and kappa angle (angle from ground east  
to image x axis) of the frame are given in degrees.  The size of the
image in pixels, rows (lines x columns), is given.  The "status" 
column indicates whether the orientation information is Complete, 
Interim, or Pending.  For images shown as "Complete", we believe 
that the orientation information is correct and adequate for RADIUS 
purposes. All motor pool images are complete, while one of the images 
of the "mountain" area (fhov725) has only GPS positions available, 
and another (fhov1525) has no orientation information available.  
			
2.  IMAGE ACQUISITION AND ORIENTATION

The imagery was flown by Photo Science, Inc., in October, 1993, using a GPS
(Global Positioning System) equipped aircraft.  A total of 5 strips of vertical
imagery were flown, 2 over the motor pool area and 3 over the mountain area.  

Photo Science was responsible for determining the image positions and 
orientations.  The vertical strips were triangulated in a simultaneous 
bundle adjustment, using field surveyed control based on GPS survey points.  
Photo-identifiable points in the imagery were measured on the vertical 
images and then used as control in single-photo resections for the oblique 
images. 

Due to a problem with PSI's single-photo resection software, the 
first set of calculated positions for the oblique images were in
error. The GPS positions were distributed as an interim solution.
PSI re-did the oblique image resections; unfortunately, for several
reasons, the results were unsatisfactory due to poor image-to-image
registration. In order to get a usable data set distributed before
the end of the project, a simultaneous adjustment of all the images
of the motor pool area (21 images total, 7 vertical
and 14 oblique) was performed.   PSI's control and pass point
measurements were used, along with additional tie points added for
geometric strength. Control points were also measured on all images on
which they were visible, and re-measured in cases where PSI's original
measurements were bad.

All image position and orientation information is distributed in the TEC header
format, with the RADIUS extensions.  A description of the TEC header format 
is included on this tape.

ASCII FBIP files for each image are also included.  Note that these
files contain some additional header fields (date, gmt,
horizontal-units, vertical-units) that are not currently supported
by RCDE. These fields can be deleted as necessary, or RCDE can
be modified to use them.

2.5. ORIENTATION PROBLEMS

As mentioned above, PSI's second attempt at calculating orientations
for the oblique images was not usable.  There are several
reasons for this, which should be kept in mind when generating future
data sets or when working with multiple oblique images.

The main problem was the use of single-image solutions, instead of one
simultaneous solution.  The lack of redundancy in a single-image
solution means that the results will be very sensitive to measurement
errors, bad control points, or image distortions.  These factors will
vary between images, so two images which separately give what appear
to be satisfactory solutions may be inconsistent with each other.

The weak geometry of oblique photos makes this problem much worse.  
The longer viewing distance, coupled with the oblique angle the
viewing rays make with the ground, means that the solution is
inherently less stable than a vertical image from the same altitude.

These factors mean that more than the theoretical minimum number of 
control points should be used; unfortunately, in most cases only four
points were used. In several cases, more control points were measured 
but apparent bad points were deleted without remeasurement, leaving 
a minimal number of points.

The control points that were chosen had a very poor distribution
across the oblique images, since most control points were surveyed
within or near the two main streets that define the motor pool area.
In the oblique images, the motor pool covers only a small band,
meaning that the control points do not cover the entire image area.
This led to high variances on the image orientations and also 
poor image-to-image registration outside the band of control points.

3. IMAGE NAMING CONVENTION

Each image name begins with the letters "fh", followed by "n" for
nadir (vertical) photography or "ov" for oblique view.  For nadir 
images, the next number is the flight number, always 7 in this case.
The last one or two numbers are the image number.  For oblique images,
the last two numbers give the date the image was taken, either 25 or
27, while the preceding one or two numbers give the image number. 
For example 'fhov1125' indicates oblique image number 11 taken on 
November 25.  The image table file gives a complete list of imagery 
organized by the image name.

4. IMAGE MEASUREMENT (.RESECT) FILES

Included with the data set are two .resect files, one (all.resect)
containing the input data to the block adjustment and 
the other (all-out.resect) containing the final coordinates and 
covariances of the ground points.

The format of the file is pretty simple:

	point -- (type of information -- always point in this case)
	<pt-id>
	2 -- (means sexagesimal coordinates)
	<N | S> <lat-degrees> <lat-minutes> <lat-seconds> <lat-thousandths of sec>
	<W | E> <lon-degrees> <lon-minutes> <lon-seconds> <lon-thousandths of sec>
	<elevation>
	<std-dev-X> <correlation XY> <correlation XZ> <std-dev-Y> <correlation YZ> <std-dev-Z>
	<n> -- (number of images point is measured on>
	<image-id> <row> <col>   (for each image) 
	...
	
The image coordinates are given as <row>, <col>, where the origin is
the upper left corner of the image, rows increase downward, and
columns increase to the right.   
***This is not the RCDE image coordinate system!!!***
The column coordinate is the same as RCDE 'v' coordinate.
To convert the row coordinate to the RCDE 'u' coordinate, subtract
it from the number of rows in the image.

PSI's point number scheme gives tie points a 4 or 6 digit number. The
6 digit numbers always begin with 90.  The first 3 digits are the
photo number on which the point appears, while the last digit is the 
point number on that photo.  Point number 2 is in the middle of the
image, while points 1 and 3 are at the edges.  So, for instance, 
point 907052 is a tie point identified on photo 705, near the middle 
of the image.  Tie points are artificial points, marked in stereo 
("pugged") on overlapping images.   Without the marked film, there is 
no way to identify the point used in the image.

Control points have 2 or 3 digit numbers, such as 94 or 102.  These 
are photo-identifiable points that have either been surveyed on the
ground or measured in the imagery to use as control for the oblique 
images.  

We also measured additional tie points between images, using 
photo-identifiable features in the images. These have arbitrary 
names, usually pass**  or t**. 

