Sensor Model Support
A core SOCET GXP® requirement that increases the capacity to generate accurate results is using data in its native format whenever possible. This extends from basic image types including TIFF, GeoTIFF, JPEG and NITF; through terrain formats such as DTED, NITF and GeoTIFF; to features and vectors such as shapefiles.
Native formats and georeferencing
Working with data in its native format has performance and usability advantages. For example, a TIFF image can be dragged from Windows Explorer into SOCET GXP for immediate viewing. If metadata exists for imagery or other data types, SOCET GXP uses that information to georeference raw data to real-world ground coordinates. The georeferencing may be as simple as a text file that identifies the data’s mapping coordinate system or a tag to extend the basic form of the raw data.
Image-to-ground references can be represented with a fundamental orthogonal projection to define the X and Y ground coordinates of a pixel in an image along with scale factors for image line and sample coordinates, or the four-corner locations of an image in ground space.
To move beyond simple planimetric XY models, GXP engineers work closely with organizations such as the Community Sensor Model (CSM) working group, government agencies, specific programs, system integrators, satellite operators, and vendors of airborne cameras, LIDAR systems and hyperspectral sensors. These relationships enable GXP engineers to develop and rigorously model the transformation between an image and the ground based on a projective mathematical function. The projective sensor model relates line and sample image coordinates to X, Y and Z object space coordinates (ground coordinates; latitude, longitude, height; easting, northing, height; etc.).
Projective sensor model
The projective sensor model relates line and sample image coordinates to X, Y and Z object space coordinates (ground coordinates; latitude, longitude, height; or easting, northing, height; etc.).
The projective model is important for SOCET GXP functionality, such as the easy-to-use height measurement and simple building tool, now available in SOCET GXP v3.2, to stereo mensuration and applications such as automatic terrain generation.
The projective model can take many forms. In some cases the model might be generic, for example, a cubic Rational Polynomial Coefficient function (RPC), frame or generic pushbroom sensor. The cubic RPC expresses image coordinates as a ratio of two cubic polynomial functions of the ground coordinates. The frame model is based on the well known collinearity equations, and the Frame-Advanced formulation used in SOCET GXP also models various systematic image errors. In the case of a generic model, the image metadata is propagated into the model components to allow an operator to measure in X, Y and Z ground space accurately. In other cases, rigorous sensor-specific models are developed, which typically rely on information about the sensor position, attitude and rate (exterior orientation) as well as focal length, chip size, chip orientation and lens characteristics (interior orientation). “Rate” is a broad term that means rate of change of position and attitude. SOCET GXP uses default selection criteria when multiple models are provided for individual images. However, users can select between generic models or rigorous models for ease of use.
Sensor models and triangulation
In some instances metadata may require triangulation to improve the accuracy of an image. The triangulation process updates the sensor’s metadata, including the adjustment of exterior and interior orientation components. These updates are determined by means of a mathematical estimation process called least squares, based on observing corresponding points, usually called “tie points” or “pass points,” in multiple images, then finding the optimal fit between these measurements and the changes being computed. SOCET GXP’s automatic point measurement function spares the operator from this tedious task, which was a massive bottleneck even a generation ago. In this scenario, the photogrammetric process is exposed, requiring subject-matter experts to perform the task, thus ensuring highly accurate imagery for further down-stream exploitation.
The SOCET GXP advantage
The benefit of using SOCET GXP for geospatial analysis and image exploitation is twofold. The application reads images natively with the associated metadata and sensor model to deliver the highest degree of accuracy through automated triangulation. In addition, complex photogrammetric procedures are simplified, making the process intuitive for novice-to-expert users, with the option to perform further calculations if desired.
As new satellites and airborne sensors become available, BAE Systems continues to implement functionality for ease-of-use, and to ensure the highest degree of accuracy as a result of core development requirements imposed on the application development process. I am excited about the SOCET GXP v3.2 release that is now shipping with a breadth of new capabilities for geospatial and image analysis.
SOCET GXP supports the following optical and radar sensors:
- ALOS (PRISM, AVNIR-2 and PALSAR sensors)
- Four-Corner: simple 2-D transformations
- Frame Advanced
- MSP 1.1.2
- Ortho: simple 2-D transformations
- RPC:STDI-0002 v3.0, Appendix E: ASDE 2.1, Section E.2.4
Figure 1. Sendai 9.0 earthquake damage of an oil refinery in Shichigahama, Japan. The red areas indicate probable oil spills that were identified using SOCET GXP v3.2 supervised classification functionality. WorldView-2 8-band imagery courtesy of DigitalGlobe.
Figure 2. This rigorous pan-sharpened image is created on-the-fly in SOCET GXP v3.2 by combining WorldView-2 panchromatic imagery at 0.6m ground sample distance (GSD) with WorldView-2 MSI imagery at 2.4m GSD. The zoomed-in view of the probable oil spill outlined in figure 2 shows rigorous projective sensor model support in the pan-sharpening process, thus allowing use of the simple height measurement tool to measure and label one of the damaged oil storage facilities.
Figure 4. Figures 3 and 4 show GeoEye-1 stereo imagery collected over Port-Au-Prince, Haiti following the January 2010 earthquake. In figure 3 displaced people occupy a soccer field. The green areas in figure 4 identify potential helicopter landing zones. SOCET GXP v3.2 Automatic Terrain Generation is used to produce a digital surface model over Port-Au-Prince with 11 million points at 3-meter spacing and a nominal GSD of 0.5m. Using the surface model, SOCET GXP locates potential helicopter landing zones with a slope of less than 5 percent and an area of 60sq meters or larger. Viewing the two images side-by-side in a SOCET GXP Multiport viewing window provides both geospatial and visual intelligence. Image courtesy of GeoEye.