Aerospace and Electronic Systems Magazine June 2017 - 30

Elimination of Resampling Errors in Wide Area Motion Imagery
six cameras, each with a different camera model. There will be
one set of camera models for the entire data set. Finally, the elevation data for the area is required. This would typically be a Digital
Terrain Elevation Database (DTED) for the area. With this information, the projection can be performed in the modified SET as
required by the analyst.

TRADITIONAL SET MODEL
Clearly there is a loss of resolution in the projection process. The
projection process in this case is a combination of resampling, and
compression. Based on the compression results above the issue is
not compression, but resampling.
In the traditional model the image is projected first. During
the projection process the image is moved from the source image
to the destination image. This normally involves a homographic
transformation for each tile of the DTED. A tile is the rectangular
area of the DTED defined by four adjacent points. The four corners
of the DTED tile are located in space by LLH, Military Grid Reference System, or other form. The location of the tile is projected
onto the raw image using the collinearity equations based on the
pose and the camera calibration. A homographic transformation
for the projection is computed based on the correspondence of the
four tile corners. The inverse homography is then used to map the
source image pixels to the destination image. The process is repeated for each tile. In this manner, the location of each destination
pixel in terms of its location in source image is found. Each destination pixel has an associated source x and y location. The locations are not integers so some interpolation method is employed.
The Intel Performance Primitives remap function [14] can be used
to perform the resampling operation.
The rectangular source image is transformed into a trapezoidal
destination image as illustrated in Figure 10. This is because the
aircraft is orbiting the center of the target area in a counter
clockwise direction and the scene is on the left side of the
aircraft. The image is captured at an angle of about 45°
with respect to the horizon as shown in Figure 10.
Because the aircraft is constantly moving, and
due to the nature of the projection process,
there is no way to use a one-to-one mapping of the source image pixels to the
destination image.
Once the image is projected
and resampled it appears as shown
in Figure 11. The image is 23,296 by
29,440. The image data set is an "R" set of
six images at progressively lower resolution.
Figure 11 shows the R0 image, the highest resolution.
The projected image is compressed and saved with the
geo-registration information as a NITF image. Figure 11 was
extracted from the NITF image for frame 280. Note the rectangular
camera images are trapezoids when projected onto the ground.
The NITF image is then sent to the viewer or SET. The SET
extracts the image and displays the image with the appropriate latitude and longitude. The SET will normally include tools to help
30

Figure 10.

Image location on ground in relation to the aircraft for six cameras.

the analyst such as zoom, drag, mark, save tracks, and save subsets
of the image. For the "R" set the most appropriate image for the
current zoom settings is displayed.

REVISED SET MODEL
Some modifications of the SET are necessary to maximize resolution. When the analyst selects a zoom that is at or above the resolution of the raw image, the area selected must be extracted from the
raw image and projected using the raw image, the pose information, and the DTED. The image must then be rendered using the
equivalent of the Matlab® pcolor function.

IEEE A&E SYSTEMS MAGAZINE

Figure 11.

Projected NITF image for frame 280.

JUNE 2017

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine June 2017