General OrthoEngine Questions

What is the recommended DEM resolution for orthorectifying an image in OrthoEngine?

When I display two adjacent ortho images in Rational Functions project, they are not precisely aligned. What can I do to improve my results?

How do I orthorectify my historical aerial photos if they do not have fiducial marks and I do not know anything about the camera the photos were taken with (i.e. the focal length)?

How do I deal with different DEM elevation references in OrthoEngine - MSL (Mean Sea Level) and height above an ellipsoid.

I want to make a DEM with a vertical resolution of 1-2 feet. What are the different variables that come into play in obtaining this?

How does the weighting of GCPs and GPS affect the block adjustment?

How do I interpret the results of my residual values? What is an acceptable residual value?

What should the order of my workflow be for my air photo project?

The GPS text file for my User Input Air Photo OrthoEngine project does not have exterior orientation information for each photo. What do I need to do to still be able to use my GPS data?

Why is my RMS error much larger in the X direction than the Y direction?

Do all my photos have to be rotated the same way for auto fiducial collection to run properly?

How does Ortho Kit imagery differ from the Geo product in terms of the number of GCPs and my accuracy expectations?

How does OrthoEngine deal with inputs that are in both feet and metre units?

How many GCPs per scene do I need to collect to ortho-recitfy satellite imagery?

I am having trouble getting the Auto Locate to work when collecting GCPs in OrthoEngine?

OrthoEngine seems to want to make an Ortho which is much larger then what it really should be given the output resolution I have set.

There are a couple of places in OrthoEngine where I have to specify a minimum and maximum elevation. How does this affect the process?

What does the Matching Threshold in Auto Tie-Point collection mean?

Do you have more information on the Rational Function Coefficients used by OrthoEngine?

Q) What is the recommended DEM resolution for orthorectifying an image in OrthoEngine?

A) The DEM resolution needed during the orthorectifiction stage in OrthoEngine depends on the resolution of the image. Ideally, a DEM with at most 10 times the image resolution should be used. For example, we recommend a DEM with a resolution of 5 or 6m for a panchromatic QuickBird scene. However, it may be difficult to obtain DEMs with such high resolutions. When no DEM is available for your area of interest, the following sources can be considered: USGS DEMs have a resolution of 30m, SRTM DEM resolution is 90m and GTOPO30 DEMs have a resolution of 1 km.

Q) When I display two adjacent ortho images in Rational Functions project, they are not precisely aligned. What can I do to improve my results?

A) This may occur if you have collected less than three GCPs in your project. If you have one or two GCPs per image, you can perform a zero-order transformation. A zero-order transformation produces a translation for x and y only. If you have at least three GCPs per image, you can perform a first-order transformation. A first-order transformation produces a translation and a rotation.

Usually performing a first-order transformation is best, except when the GCPs are not well distributed. If your GCPs are clustered together, a first-order transformation may introduce new and significant errors in the image away from the GCPs. If your GCPs are not well distributed, you will probably obtain better results from the zero-order transformation. For example, if you have a project with no GCPs and a number of accurate tie points, you can improve the model by collecting tie points and using a zero order transformation.

You can change the adjustment order in the project in the GCP collection window.

Q) How do I orthorectify my historical aerial photos if they do not have fiducial marks and I do not know anything about the camera the photos were taken with (i.e. the focal length)?

A)

1) If you have fiducials but NO coordinates:

Measure between each set of fiducials along the edge of the photo, in mm. Use the "Compute from Length" button and enter the four lengths.

2) If you have NO fiducials but do you have a cross marking of the photo center:

Measure the distance from photo center to each edge, in mm. Use the corner pixels of the photo as fiducials and enter the following coordinates for the camera calibration information as fiducial coordinates:

Top Left: X=-X1 Y=Y1

Top Right: X=X2 Y=Y1

Bottom Right: X=X2 Y=-Y2

Bottom Left: X=-X1 Y=-Y2

3) If you have NO fiducials, and NO photo center:

Measure the length of each side of the photo using "Compute from Length" in the Camera Calibration panel - "Fiducial Coordinates" section. Enter your measurements, in mm. Use the image corners as the fiducials.

4) Focal length of the camera is unknown:

If the focal length of the camera is unknown, there are a few different things you can try:

1) Most older cameras were constructed with 6", 8" or 12" focal lengths. A first approach would be to try each of these. If you do not have the right focal length, the resulting flying height will be way off, and you will have remaining distortions in the imagery.

2) Publications such as the "ASPRS Manual of Photogrammetry" list photogrammetric cameras through history, with frame sizes and nominal focal lengths. If you have an unusual frame size i.e., not 9"x 9", this would be a valuable tool.

Note: The focal length estimates need only be performed for one photo. After that, you can simply enter the same data for the remainder of the photos.

Q) How do I deal with different DEM elevation references in OrthoEngine - MSL (Mean Sea Level) and height above an ellipsoid.

A) Satellite Orbital Modelling - Toutin's Model , Low Resolution AVHRR and Aerial Photography projects use the MSL datum.

- If you enter the GCP elevation values directly, the elevation is assumed to be MSL.

- If your GCP elevations are extracted from a DEM and the DEM uses ellipsoidal heights, you can use Utilities>Convert DEM Datum to convert the DEM to MSL. Alternatively, you can specify the ellipsoidal datum when you select the DEM and the elevation values will be converted to MSL "on-the-fly". The same applies when selecting the DEM in the ortho image production processing step.

- If you use an elevation offset in the ortho image production step, the elevation values are assumed to be above MSL.

Note that if you extract a DEM from stereo images in any of these types of OrthoEngine projects, the DEM produced will use MSL

For Rational Functions and Satellite Orbital Modelling - RADARSAT Specific Model projects, elevations can be either MSL or ellipsoidal.

- If you enter the GCP elevations directly, you can specify the elevation datum using Options>GCP Elevation Datum before you collect your GCP's.

- If your GCP elevations are extracted from a DEM, you need to specify the elevation datum when you select the DEM. In this case, the elevation values will be converted to height above ellipsoid "on-the-fly". The same applies when selecting the DEM in the ortho image production processing step.

- If you use an elevation offset in the ortho image production step, the elevation is assumed to be height above ellipsoid.

Note that if you extract a DEM from stereo images in any of these types of OrthoEngine projects, the DEM produced will use height above ellipsoid.

Q) I want to make a DEM with a vertical resolution of 1-2 feet. What are the different variables that come into play in obtaining this?

A) 1. Generally speaking, from traditional film based aerial photography (9 inch format) you would expect accuracies of approximately 30-40 microns in the vertical direction. In other words, for 1:10,000 scale photography you would expect 30-40cm residuals for your bundle adjustment (1:5,000 -> 15-20cm; 1:20,000 -> 60-80cm; and so on). To obtain a DEM that is suitable for 1 or 2 foot contour intervals you would aim for a better ground accuracy (maybe better than half the desired vertical resolution -> 15 cm or 30 cm) so that it will accurately represent 1 or 2 foot deviations in the terrain. The flying height that this corresponds to for a standard aerial camera and different scales are: 1:10,000 -> 5,000 feet above ground level; 1:5,000 -> 2,500 feet AGL; 1:20,000 -> 10,000 feet; and so on.

2. As for angular separation of the two images, it is most efficient to collect vertical photography with a base to height ratio of 0.6 to optimize the geometry of the intersecting rays from which the DEM is derived. With a base to height ratio of 0.6 for vertical photography the resultant images will be overlapping by 60% and if taken from 10,000 feet AGL they will be taken 6,000 feet apart.

3. The resultant accuracy of the DEM is largely dependent on the ground sampling distance and the terrain variability. Clearly, if you only extracted elevation values every 10 meters it wouldn't capture 1 foot undulations very well in between samples, unless of course the terrain was flat. Unfortunately, there is not a hard and fast rule for this. The sampling interval should be chosen appropriately for each particular scene (large variations in terrain, flat ground, breaklines collected, etc.).

4. Scanning resolution should also be considered. Of course, this is one of those unanswerable questions, right? Technically, if you wish to create a DEM with the vertical resolution, you also need to consider the accuracy of the model i.e. if your DEM requires to be sub-foot accurate in Z, then you must make sure it is sub-foot accurate in XY, because offsets in XY will mean your Z is also incorrect at any point.

5. The DEM extraction itself has many options for down sampling, pixel spacing, DEM detail, which may need to be considered.

Q) How does the weighting of GCPs and GPS affect the block adjustment?

A) The weighting of GCPs and GPS affects the influence of those values when they are used in the block adjustment. They are weighted inversely proportionately to the error value, so GCPs or photo centres with high error will affect the solution less than points with lower error. How you decide what values to use generally depends on what you want to accomplish. Usually, the first step is to put what you believe to be reasonable values in for both the GPS data and the GCPs. For the GPS, you can get an estimate from the manufacturer. For the GCPs, you have to look at the source and estimate their accuracy. After you have your initial estimates, you can always change the values to force one or the other (GPS or GCPs) to have the dominant effect on the solution by setting those errors closer to zero.

Q) How do I interpret the results of my residual values? What is an acceptable residual value?

A) Residuals are the difference between where the GCP has been placed, and where the bundle adjustment computes the position of that point to be. The bundle adjustment solves for the best possible solution for the location of each image, using all GCPs, TPs, and EO (exterior orientation) data available. The criterion for this solution is that the sum of the square errors is minimized (least squares), which means that no single GCP or TP will fit perfectly. The residuals are the remaining shift in the computed position. Residuals are not errors to be corrected, they simply help you to see which GCPs fit well (low residuals) and which don't fit quite as well (higher residuals). High residuals can tell you if a single GCP is severely out of place, or more often, where parts of the block of imagery do not fit the ground well.

Normally, you aim for residuals of less than 1 pixel. This assumes that you have the ability to measure the GCPs to that accuracy on the imagery. However, if your GCPs are not accurate to 1 image pixel (often the case for high resolution imagery), then it is unlikely that you will achieve that goal. If you are using GPS/INS, you will be limited to the accuracy of that data. Also, if you are using a DEM for GCP elevations, this can introduce error. So, the fact that you don't achieve residuals under 1 pixel does not necessarily mean you do not have the best possible solution. The magnitude of the residuals should reflect the accuracy of all data sources, as well as the image resolution.

Data snooping refers to standardized residuals, which are the residual values divided by the RMS error. This allows you to tell if your residuals are abnormally high, or just appear high because the overall accuracy is low. For example, you may have residuals of 10m. You find, however, that your GCPs are only accurate to 20m. This produces a high RMS error for the block. When the residuals are normalized, you may find values such as 0.002, indicating that the residual is probably OK for the variability of that data. I find that data snooping is valuable in cases where you are trying to determine how accurate your GCP or EO data is.

Q) What should the order of my workflow be for my air photo project?

A)

a) Create the project and import the imagery
b) Import the GPS/INS data for all available photos
c) Enter GCPs, if desired.
d) Enter at LEAST Tie Points (manually) and GCPs if possible, for any photos which do not have GPS/INS. The auto tie point can only operate if there is an initial approximation for where each photo is. Otherwise, it won't know what photos to match with.
e) Run auto tie point.

Q) The GPS text file for my User Input air photo OrthoEngine project does not have exterior Orientation Information for each photo. What do I need to do to still be able to use my GPS data?

A) You can use photos with no EO values, but they must be tied into the block using tie points in order to obtain an approximate solution first. If available, also collect GCPs for these photos.

Q) Why is my RMS error much larger in the X direction than the Y direction?

A) This is caused by errors in your DEM. For a side-looking sensors, pixel displacement (parallax) due to terrain elevation is mostly in the along the line direction. Consequently, the errors in DEM have much stronger effect on the error in X direction. A 3:1 ratio between X and Y errors seems quite reasonable, particularly for a satellite sensor with good attitude control.

Q) Do all my photos have to be rotated the same way for auto fiducial collection to run properly?

A) The auto fiducial collection will run successfully for any combination of orientations. However, the calibration edge has to be set manually. There is no automatic way to do this, since the digital image can be in any orientation. So, you have two options:

a) Run the automatic fiducial collection on the raw images, and then manually change the calibration edge for the affected photos. or

b) Rotate the images to a common orientation using the batch rotate function before running the auto fiducial collection. The calibration edge does not need to be adjusted in this case.

The first option is generally the best as it is easiest to work with photos if they are all scanned with left to right overlap, with the up axis of the digital images all pointing the same way (i.e all North-up, or all East-Up, or all 45-degrees-Up.) That makes stereo viewing possible, makes epipolar generation easier, and makes the GCP collection more intuitive.

Q) How does Ortho Kit imagery differ from the Geo product in terms of the number of GCPs and my accuracy expectations?

A) The Geo product imagery is less nadir than the Ortho Kit imagery. This means that the Geo product is also more sensitive to the DEM accuracy than the Ortho Kit product. Therefore it requires more GCPs to get an accurate orthoimage than the Ortho Kit data, because with the Geo product the sensor model is derived from ground control. However, because the Geo product data is less nadir it may be better for DEM generation than the Ortho Kit data---researchers are still investigating DEM results with both products The Geo product is also cheaper to purchase than the Ortho Kit product.

Ortho Kit requires fewer GCPs because it uses GCPs in conjunction with the Image Geometry Model (rational functions) information. The number of GCPs is not view angle related.

Q) How does OrthoEngine deal with inputs that are in both feet and metre units?

A) There are a number of areas within OrthoEngine where units could be represented in feet or metres. You should carefully consider the units during the course of your project.

1) Units for the elevation of the input GCP's
2) Units for the elevation of the input DEM. May be the same as GCP's or may not be.
3) Units for the Output projection (i.e. FOOT or SPAF)
4) Units for the GCP projection (i.e. FOOT or SPAF)

How this fits into your OrthoEngine workflow:

1) Options > Elevation Units: This is set based on the elevation of the GCPs imported from a text file, or the altitude as given in an exterior orientation file. If you are extracting GCP elevations from your DEM, then use the same units of elevation as your DEM. Default is meter.

2) DEM units for orthorectification. If the elevation units of your DEM do not match the units of the Output Projection, then the "Elevation Scale" setting (under DEM options in the Ortho Generation panel) must be set to scale the data to the correct units. Alternatively, the "Utilities" > "Replace Image Values" tool can be used to transform the elevations to the correct units. For example,if the output projection is in a foot-based projection (FOOT or SPAF) then the DEM elevation units must be in feet. If the output projection is a metre based projection (UTM, SPCS) then the DEM elevation units must be in metre (feet>metre multiply 0.3048, metre>feet divide 0.3048)

Version 10 provides an option within the Ortho Generation panel to select the appropriate units so that elevations of either feet or meters can be used with any output project projection.

3) Automatic Tie Point Average Elevation: The average elevation should be set based on the GCP Elevation Units, set under the Options menu.

4) DEM Generation Elevation Range: These values should be set based on the units of the output projection. (i.e. feet for FOOT or SPAF, metres for SPCS or UTM).

Q) How many GCPs per scene do I need to collect to ortho-recitfy satellite imagery?

A) Minimum number of GCPs under ideal conditions

ASAR (Envisat - 1B) 8
ASTER (1A or 1B) 6 (6 - 8 recommended)
CBERS 6
EOC 6 (6 - 8 recommended)
EROS (1A) 8
ERS/RADARSAT 6
FOMOSAT 6
IKONOS Geo Product 6
IRS 6 (6 - 8 recommended)
JERS1 8
Landsat 6 (10 - 12 recommended)
MERIS (Envisat - 1B) 6
ORBVIEW 6
RADARSAT Specific Model - none (accuracy improved with 1 - 2)
RADARSAT Toutin's Model 8
SPOT 1, 2, 3, 4 4
SPOT 5 (1A) 6
QuickBird (Basic) 6
QuickBird (Ortho Ready Standard) 8
Rational Functions Computed from GCPs: • 5 per image (19 per image is
recommended)
Rational Functions Extracted from ImageFile: None required (accuracy is improved
with 1 or more GCPs)

In practice it is best to gather double the number of GCPs, as GCPs will contain errors. Click here for more details on the Sensor Models.

Q) I am having trouble getting the Auto Locate to work when collecting GCPs in Orthoengine?

A) The following steps explain how to use the Auto Locate function:

1) Load your uncorrected image, your GCP source and your DEM file.
2) Make sure the Auto Locate button is toggled on in the GCP collection panel.
3) Select a GCP on you Georeferenced image, hit the Use Point button at the top of the Georeferenced image.
4) Hit the Extract Elevation button on the GCP collection panel. The cursor should move to the proper location on the uncorrected image.
5) Hit the Use Point button on the uncorrected image.
6) Accept the GCP in the GCP Collection Panel.
7) Follow the above instruction for your next point.

Q) OrthoEngine seems to want to make an Ortho which is much larger then what it really should be given the output resolution I have set.

A) OrthoEngine automatically minimizes the size of the ortho file by calculating the output footprint of the ortho prior to creating the file. However, if the ortho footprint overlaps background data in your DEM (such as -150 or -999999), the apparent footprint of the ortho becomes very large. This is the cause of large black images with the orthos in the center. Try setting the background value for your DEM, and then allow the software to recompute the footprints. They should be much smaller, trimmed close to the edges of your ortho data. To find out the DEM background value, open the DEM in an image viewer. Click into the black areas in the DEM, and look at the values shown in the bottom of the OrthoEngine panel. It will display the X, Y coordinates, and the "Int:" value. This intensity is the background elevation. When you use the "Select" button on the ortho panel, you will see a text box for background elevation.

Q) There are a couple of places in OrthoEngine where I have to specify a minimum and maximum elevation. How does this affect the process?

A) The approximate elevation helps the matching process by giving it a good first guess at the parallax that it should find between the images. The matching is based on a small tile (area) of imagery. If you don't give an estimated elevation, it has to assume an elevation of 0. If your ground elevation is significantly different, then the corresponding features from the right and left images might not appear in both of the small tiles, since the location has not been adjusted for the approximate offset due to elevation. So, setting the approximate elevation improves the success rate of the matching, and reduces blunders.

Q) What does the Matching Threshold in Auto Tie-Point collection mean?

A) Matching Threshold is the minimum correlation score that will be considered to be a valid match. Setting the matching threshold adjusts the quality statistic so that only the best correlations are considered successful matches, or so that less successful correlations are accepted. This will not affect the speed, but rather the total number of points that you get. For example, if you have really clear, high-resolution data, you may be able to set your matching threshold very high, to 0.85 or better. This will make sure that you get very highly correlated points, and won't accept any questionable matches. However, if you are dealing with lower resolution, or fuzzier data, you are likely to get very few successful matches with a threshold of 0.75 or better. In this case, you would change it to a lower value such as 0.6 in order to accept more points.

Q) Do you have more information on the Rational Function Coefficients used by Orthoengine?

A) Rational Functions (RF) are commonly used in satellite imagery. It is a way to supply the ortho transformation coefficients without releasing the sensor information. It is an approximation of a rigorous model used by NIMA for their NITF format, Space Imaging for their IKONOS data and Digital Globe for QuickBird imagery

RF are better than standard polynomials because it supports elevation as well. Thin Plate Spline (TPS) requires many more GCPs than RF. For example, 20 coefficients require only 40 GCPs for RF. TPS is like an exact fit (each point has a very small area of influence) and it works well if you have many GCPs.

Online Product Support

	Tutorials
	FAQs
	Modeler/Script Exchange
	Contact Support