Automatic Tie Pointing
Contents How do we densify a block of measurements? What is the best way to thin data? How do we deal with EO’s in one coordinate system and control in another? What is the fastest/best way to run a strip job? Can blocks be run in sections? How do we get more multi-ray (cross-strip) points? How do we know if the points are good? Auto Tie Given Image Positions (XyzTie) Automatic_Tie_Points_Densification_(AutDen) Semi-Automatic Auto Tie (SemTie) Clear Exterior Orientations (CleEos) Delete Low Ray Points (DelLow)
The workflows below are largely the same. Their performance is, in large part, determined by the input data. The project wizard is designed to help make this happen. The "Orientations and Control" tab's control thumb nail viewing is an excellent to check imported EO and control data. If EO data has been imported, it is wise to measure control before running auto tie. The AT layout can be used to visually confirm strip and photo numbers. In short, it is critical to get the inputs correct before running auto tie. It is frequently better to leave data out than use poor or erroneous data.
How do we densify a block of measurements? Some users want to use AutoTie only after they have minimally tied a block of images manually. The Auto Densify (AutDen) was written for them. It does not look for any new overlaps, but adds more data to those that are already known. That is if imageA and imageB already have some ties between them then it will attempt to add more. If EO’s have been imported (see ImpGps) this process will also extend matches into new images.
What is the best way to thin data? Least Squares match (LsMat). The LsMat command can be used to thin data, refine data, and push measurements through to more images (if EO’s have been imported).
How do we deal with EO’s in one coordinate system and control in another? Auto tie is designed to work in any rectangular coordinate system. So the approach taken is to do the auto tie in one system and then switch to the other.
What is the fastest/best way to run a strip job? This answer depends, but the short version is: Just like any other block. If you have EOs then use the work flow given EOs. If you don’t you must follow the work flow without EOs, which requires the user to provide some ties between the strips. In the special case of single strip job the SemTie work flow is frequently the fastest whether you have EOs or not. This is counterintuitive, but to run SemTie does not require EOs to be imported or even a correct camera file. Further on a single strip no manual measures are required. Hence, as soon as strip and photo numbers can be assigned it can run—users can worry about importing EOs, setting up cameras, etc. later if at all.
Can blocks be run in sections? Yes, the Auto Tie Parameter ‘Min Auto Tie Sequence number’ allows user to start auto tie numbering arbitrarily and avoid duplicate point names in sub-blocks. For example: In a 30 strip job, strips 1-11 could be run with a ‘Min Auto Tie Sequence number’ of 0, strips 11-21 with a ‘Min Auto Tie Sequence number’ of 1000000, and strips 21-30 with a ‘Min Auto Tie Sequence number’ of 2000000 (If there might be more than a million auto ties in a sub-block use additional zeros). Care must be taken that the ATM files of the separate runs do not overwrite each other. The surest way to do this is to set up different projects (IN DIFFERENT DIRECTORIES) that reference the sub sets of images, but it can also be accomplished with the BacAtm command. The ATM files can then even be combined for a final bundle.
How do we get more multi-ray (cross-strip) points? The best way to do this is to get lots of along strip matches and reasonable EO’s then use LsMat to push the along strip matches through to the other images. The best procedure depends on what data is already available. For instance:
How do we know if the points are good? Statistics and residuals are great tools (as in bundle residual reports), but in addition looking at a sample of the points and checking their distribution is highly recommended. When reviewing points, choose cross strip points scattered around the project to get a feel for their quality. Running a free net is also a good idea. It will also reveal any problems in the block more reliably than a bundle. A few strategically added manual points can also help gauge the quality of a block—if the bundle wants to flag the manual points as blunders there may be an issue with the auto ties. In short, be conscientious, particularly when dealing with forests and jungles. If a block would be hard for a human it will be hard for a computer.
Each of the automatic tie point measurement methods is built primarily on feature matching technology similar to the now famous "Scale Invariant Feature Transform" or SIFT. Winnowing of measurements is done using RANSAC type robust analysis. Feature matches are optionally refined (and further winnowed) using an optimized version of Gruen's adaptive least squares matcher. There is also some limited use of normalized cross correlation when doing point propagation.
The various commands below are designed to function with different amounts of input data. 'AutTie' is the general command that will automatically pick a method based on the available data.
The parameters in the table below are all of those that are used in the various methods. Every method does no use every parameter.
Auto tie is the general command that will automatically run the most powerful method it can based on the available data. This behavior can be manually overridden by keying in the command for a specific auto tie method (e.g. BruTie, FasTie, SemTie, OblTie or XyzTie). This command can be executed from the drop down menu or by using the key in "AutTie". Thus, for many data sets it is never necessary to open layout.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
See the auto tie parameter list here.
Auto Tie Given Image Positions (XyzTie) XyzTie is specialized for tying nadir looking images with measured positions. The "AutTie" command (from the drop down menu or the key in) will also execute this command if the images all have measured positions by do not all have measured orientations.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
See the auto tie parameter list here.
Automatic Tie Point Densification (AutDen) When executed from the key in (AutDen) or from the layout Adjust menu auto densify attempts to automatically densify tie points among all the images open in layout. It is also possible to run auto densify on smaller set of images by using the version available on the layout add points menu. This process differs from auto tie pointing because it does not attempt to detect overlap among images; it adds more points among images that already have some points connecting them. Thus, if two images have no points measured between them auto densify will not add any.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
See the auto tie parameter list here.
BruTie is the last resort auto tie method. It is the least reliable and the slowest method, however, it does not require EOs, strip and photo numbers, or camera data. This command will not be started by the "AutTie" command. It must be keyed in.
See the auto tie parameter list here.
FasTie is the preferred auto tie method for images with EOs. The "AutTie" command (from the drop down menu or the key in) will also execute this command if the images all have measured EOs.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
See the auto tie parameter list here.
This command is an alternative user override to the typical work flow with EOs that is intended for oblique images. A comprehensive set of feature matching among the images is done. Disabling least squares matching is recommend for oblique images. The "AutTie" command will not run this method; it must be keyed in.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
See the auto tie parameter list here.
Semi-Automatic Auto Tie (SemTie) This command is part of the auto work flow without EOs. Strip and photo numbers are required for this method. Images are automatically linked along strips. A limited number of prior cross strips ties are needed as seeds for connecting the strips.
The user is prompted for bundle parameters when executing this command because the bundle is used to winnow and thin measurements at various stages of the process.
Copies all the ATM files (point measurements) to a user specified directory. The can be restored with the restore ATM command (ResATM).
Residual information in pixels is available for every point in the image network after running a VrAdjust bundle. This command allows the user to quickly delete all auto ties (image points with names beginning with 'AT') which have residuals above the user defined threshold. A single thresholding value is used, and is compared to the magnitude of the total residual vector. For example, every point has an x residual (Vx) and a y residual (Vy); the total residual magnitude is sqrt(Vx*Vx + Vy*Vy). To assist with choosing a threshold percentiles of the total residual magnitudes are provide in the bundle residual reports; it is recommend to choose a value between the 95th and 99th percentile.
Clear Exterior Orientations (CleEos) This command clears all imported EO data. It is sometimes necessary to unload previously imported exterior orientations so that the program will not consider them in during the bundle adjustment or auto tie.
Delete all Auto Ties in a project
Delete Low Ray Count Points (DelLow) This command quickly deletes points with less than a certain ray count. This can be useful for thinning data, but is most often used to delete 2-ray points. 2-ray points have the geometric weakness that all error in depth (errors along epipolar lines) is undetectable. Hence, if the bundle points are going to be used for a surface (e.g. for orthographic resampling) it is sometimes better to remove the two ray points first.
Does patch based least squares matching to refine all the points currently in the block. It will overwrite auto ties, but creates duplicates of other points to refine. If EO's have been defined with import GPS or bundle has been run the measurements with also be extended into new images. The maximum pts per image parameter lets the user do optional point thinning as well. Note that prior to LS matching a bundle is done to refine the imported EOs.
Restores a set of saved ATM files.
|