Professional Documents
Culture Documents
A rational polynomial coefficient (RPC) model is a type of sensor model, which is a mathematical model that maps the physical relationship between 3D ground points and 2D image points. RPCs are not the same as a map projection. Most modern highresolution sensors such as QuickBird and GeoEye-1 include pre-computed RPCs with their imagery. If your file has RPCs, you can use them to georeference image pixels. In this tutorial, you will use ground control points (GCPs) and a digital elevation model (DEM) to improve the accuracy of RPC orthorectification for a QuickBird scene of El Paso, Texas.
Description
DEM\ GCPs\
This folder contains a multispectral QuickBird image of El Paso, TX (in GeoTIFF format), at 2.44 meter resolution, captured on January 5, 2002. The folder also includes associated ephemeris data, metadata, and RPCs. This folder contains a National Elevation Dataset (NED) digital elevation model at 1/3 arc second resolution. This folder contains a shapefile with 21 GCPs that lie within the extent of the El Paso image.
Page 1 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
6. Click Next. The image displays, and the RPC Refinement panel appears. A new view is added to the Layer Manager that contains an Overview, GCPs, and the image filename. 7. From the Zoom To drop-down list in the main toolbar, select Full Extent.
Page 2 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
The green check marks indicate that these GCPs are active and will be used to adjust the RPC model. These points are called adjustment GCPs. When you select a GCP in the list, the details of the GCP location appear in the GCP Properties table to the right. Included are the longitude and latitude in a Geographic Lat/Lon WGS-84 projection; height in meters above the WGS-84 ellipsoid; image coordinates; status (adjustment or independent); and error statistics. The next section will discuss the error statistics in more detail.
Page 3 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
The Horizontal Accuracy value is the horizontal root mean square error (RMSE) in meters. In the example above, all of the adjustment GCPs taken together would yield an orthorectified image that is accurate to within 33 meters. You will try to reduce this value by assessing the quality of your GCPs.
3. Select GCP 1 from the list, The image display centers over that point, and the GCP 1 marker is surrounded by a cyan-colored box. 4. Right-click on GCP 1 in the list, and select Change GCPstatus. It changes from an adjustment GCP to an independent GCP (indicated by a grey diamond symbol). It remains in the GCP list and provides an independent accuracy measurement, but it will not be used to adjust the RPC model and will not contribute to the overall model error. Changing the GCPstatus allows you to turn GCPs on and off without actually deleting them. 5. Right-click on GCP 1 again and select Change GCP status to change it back to an adjustment GCP.
Viewing Errors
In this section, you will assess the quality of your GCPs and how they contribute to the overall accuracy of the orthorectification, Once you identify GCPs with large errors, you can adjust them or delete them, then re-evaluate the accuracy of the orthorectification. Rather than give a mathematical definition of all of the available error statistics, this tutorial will highlight the most important ones to evaluate and give you a more practical approach to understanding their meaning. For more details on the error statistics, click the Help button in the lower-left corner of the RPC Refinement panel. When the ENVI Help displays, follow the link to "Accuracy Assessment Background." 1. In the RPC Refinement panel, click the Show error overlay button . A transparent image overlays the scene, showing a color gradient of interpolated GCP errors. The darkest areas indicate GCPs with the least error, and the bright red areas indicate GCPs with higher errors. The white area represents a potentially large error with a GCP location. If you scroll to the middle of the GCP list, you will see that GCP 14 has a green X marker. This means that the difference between the image-estimated ground coordinate and the GCP location is greater than three times the root mean square error in the X or Y direction. The error overlay should look like the following image:
Page 4 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
The error overlay is useful for quickly evaluating how the GCP errors are distributed across the entire image. Use the Zoom and Pan tools in the main toolbar to explore the image in more detail if needed. 2. In the RPC Refinement panel, select GCP 14. The image display centers over that GCPlocation. The GCP marker is surrounded by a cyan-colored box in the image display. 3. Look at the Error Magnitude value in the GCP Properties table. Compare its value of 137.20 meters to the Error Magnitude values of other GCPs in the list, and you should see that this GCP does have the highest error.
Page 5 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
4. From the Zoom To drop-down list in the main toolbar, select 1:1,000. 5. In the RPC Refinement panel, click the Hide error overlay button the error overlay. to turn off
6. In the RPC Refinement panel, click the Show error vector button . This tool provides another way to view individual GCP errors. You should see something similar to the following:
The cyan-colored vector shows the direction and the magnitude of the error in GCP 14. Notice how the vector points to the southwest. If you look at the Error X value in the GCP Properties table, you will see that GCP 14 shows a predicted difference of -20.98 meters between the ground point and the corresponding image point, in the x-direction. Since the value is negative, the error is to the west. The Error Y value shows a predicted difference of -135.59 meters in the y-direction; since the value is negative, the error is to the south. So both values confirm what the error vector shows.
Note: The Error Z value is not shown in the error vector. It represents a predicted difference between the Height value of the GCP and the corresponding DEM height above the WGS-84 ellipsoid.
The length of the cyan-colored vector corresponds to the Error Magnitude value (137.2 meters).
Page 6 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
The red circle around GCP14 shows the circular standard error at the 95% confidence level.This means that after you orthorectify the image, you are 95% confident that the error of this point will be within this circle. The CE95 value under the Statistics tab gives the numeric value of this circle, in meters.
Correcting Errors
If you have more accurate geolocation information available for a given GCP, you can edit its Map X and Map Y values in the GCP Properties table. You can also experiment with different Image X and Image Y values. However, in most cases, it is more practical to interactively move the GCPlocation in the image display, then reevaluate its error statistics. 1. With the Select icon active in the main toolbar, click and drag the GCP 14 marker southeast to the starting point of the cyan-colored error vector.
The error vector and error overlay both automatically update. You can see from the updated error vector and the Error Magnitude value how the error decreased for this GCP. But how did this affect the overall model? How did the Horizontal Accuracy value change? 2. From the Zoom To drop-down list, select Full Extent. Notice how the GCP errors have been re-computed and how this affects the error overlay.
Page 7 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
3. In the RPC Refinement panel, select GCP 14 and click the Delete button . Interactively moving the GCPlocation decreased its error, but after visually evaluating this GCP, you realize the position and elevation are incorrect. You do not have much confidence in the position of the GCP based on accurate site maps or aerial photos, so you choose to delete the GCP. The Horizontal Accuracy is now 5 meters. 4. You can continue selecting GCPs and interactively moving their locations, but the entire model will recompute each time and you will still see areas with high errors when viewing the error overlay. This process could go on indefinitely. You do not want to adjust too many GCP locations; two or three will suffice.
2. Select the Statistics tab. The RMSE Z value is the root mean square of the Error Z values for the entire image. It reports the vertical accuracy of the entire RPC model. The LE95 value is the difference (in meters) between the GCPmeasured elevation and the DEM elevation, using the calculated geoid offset at the 95% confidence level.
Page 8 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.
If you quit the workflow, you can continue where you left off by re-loading the input image, DEM, and new GCPshapefile. Now that you have reduced the individual GCP errors and the overall model error, you can preview the orthorectification in a limited area before orthorectifying the entire image.
Page 9 of 9 RPCOrthorectification Tutorial 2013Exelis Visual Information Solutions, Inc. All Rights Reserved.