Adjust slave image

Use case
Description of the process
Adjust slave image dialog
Adjust slave image report

Use case

• Metria CityModeler is used to make measurements in a stereo pair.
• A new stereo model is created by replacing one of the images in the original stereo pair with a new image.
• The measured objects are correctly projected to the original image, but not on the new image.
• Click the "Adjust RPC" button in the control panel to adjust the slave image and follow the instructions in the dialog. The sensor parameters in the new ("slave") image are adjusted.
• Now, any combination of images can be used to form a stereo pair and the measured objects are correctly placed in all of them.

You can run "Adjust slave image" without any previous measurements, in which case you should get a relatively (but not absolutetly) oriented model. It is however more likely that the process will succeed if you have approximatelly ten or more measured objects in the first model and these objects are also visible in the new image of the second model.

Description of the process

• If the target and/or search image is multispectral, it is converted to 16-bit grayscale images.
• All measured objects are matched. The positions in object space and in the master image are fixed, whereas the position in the slave image is determined by image matching. The points to match are chosen slightly inside the measured polygons to reduce problems with matching at occlussions. The sensor model for the slave image is used to define an approximate search area, but matching is otherwise not conditioned on the current sensor model. Only strong matching candidates are kept, defined by a cross correlation coefficient above 0.8. Matching is made with sub pixel accuracy.
• If less than five points from the measured objects are successfully matched, you are asked if you want to continue anyway. If you choose Yes, you should be aware that the estimation may fail. The estimation that will take place will not use the measured objects at all, and will thus not make any absolute orientation. A relative orientation will be done based on the gridded tie points, but since a robust translation based on the ground control points (see next bullet) will not be done, it is less likely that it will succeed. If the process is not successful, you should evaluate if model and/or object adjustments can be used to create a proper stereo model.
• Two global translations are determined by taking the median of the differences between matched image positions and estimated with the existing sensor model. Gross errors are iteratively removed if their residuals are larger than three times the standard deviations. The sensor model is updated with the estimated translations.
• A grid of 40 x 40 points is defined spread over the area of the slave image. Within an area around each grid point the most texture rich area is identified to avoid matching at homogeneous areas. Each grid point is matched with the master image to get pairs, (x',y',x'',y''), of matching image coordinates. Only matches with a correlation greater than 0.8 are kept.
• Sensor parameters are computed with least-squares estimation from observations of both the matches of originally measured objects and the matched grid points. Since we need the sensor parameters from both images to determine the object coordinates of the grid points, we iteratively first estimate the object coordinates with the last computed sensor parameters, then recompute the sensor parameters. The original sensor model, corrected for the initially estimated translations, is used as fictive observations to make sure that the parameters do not diverge too much from them.

If there are no original measured objects or if matching failed for the measured objects, the first two bullets are not executed. In general, the process works fine also when using only the grid tie points. The result is however only a relative orientation, i.e. you will be able to bring measured objects to coincide, but the object coordinates may need to be translated and rescaled.

Adjust slave image dialog

Figure 10.1: Dialog for slave image adjustment.

• Image to adjust. Choose which image you want to re-estimate the sensor model for. The other image will keep its sensor model intact. To help you make the correct decision, the image file names are written next to the Top and Bottom options.
• Information. A brief explanation of the available options.
• Image navigation. Navigate in the images with these buttons. This is helpful to view the results before accepting or declining the RPC adjustments.
• Apply parameter estimation. Starts the estimation process. When started, the progress is displayed in a progress bar together with a descriptive text. Note, that identical operations are performed on the overview images first and then on the original images so you will see each operation (e.g. "Matching points..." as shown in the dialog above) twice before the estimation ends. Once the estimation is done, your default web browser will be launched with a report, which summarizes the results. The images in the main image dialog are refreshed to reflect the new sensor model, so you can also visually inspect the part of the model, that is currently shown under the dialog, to verify that the estimated sensor model is correct.
• Ok. Ok is disabled until you have applied the sensor estimation. After that, you approve and permanently apply the estimated sensor model by clicking Ok. You will get a message which tells where the previous sensor model was backed up. If you at a later point want to undo the sensor estimation, you remove the file with the estimated sensor model and rename the backed up file to the original sensor file name.
• Cancel If you choose Cancel after you have applied the sensor estimation, the original sensor model will be reset and the image view will be refreshed so that no trace of the sensor estimation is left.

Adjust slave image report

The document is saved in the same location as the slave image and will have the same name as the image, but with extension html and with the letters "_se" (for "sensor estimation")added to the file name. However, if the directory is write-protected, then the report will be saved as a temporary file. If you would like to keep the report and it is on a temporary file, then you should use the web browser and save it in another location.

The report has some fields in green or red to indicate what the system finds correct or suspicious. They are however only hints on what to look at - the process may have succeeded inspite of red fileds and it may have failed even if there are no red fields.

The report contains the following sections, first for processing on overview images and then for processing on original images:

• Parameters. Which parameters were used in the process.
• Matching of existing object. The number of objects, the number of points, and the number of matched points.
• Preliminary translations. How much did the sensor model is moved in image space, the standard deviation before and after translations, and the number of used points before and after gross error detection.
• Matching of tie points. The number of grid points where matching is done and the number of successful matches.
• Parameter estimation. Standard deviation and the number of observations after gross error detection. Verify that standard deviation is below 5 pixels after matching in the original images.
• Transform parameters. Parameters before and after processing.