AT Setting Instruction
Last updated
Last updated
Objective of Aerial Triangulation:
Aerial Triangulation (AT) aims to use the camera's imaging model to recover the camera's position and orientation during capture.
Relative Orientation utilizes the camera's initial internal orientation elements and the relationship between correspondence points to recover the camera’s position and orientation in the relative coordinate system, along with camera’s internal parameters, distortion, and object space point coordinate.
Absolute Orientation converts the relative local coordinate system into the absolute world coordinate system using results from the relative orientation and ground control points.
Relative Orientation: Input photos (POS optional), recover the photo's position and orientation, and generate a sparse point cloud.
Absolute Orientation: Add control points to the relative AT results, perform absolute orientation, and improve the absolute accuracy of AT.
Custom: Custom AT settings for specific AT scenarios.
More: Additional AT settings. The interface simplifies default operations and opens up more AT parameters, allowing users to adjust settings to better suit the current scenario when processing data.
Both Bundle Constraint and Positioning Method can be selected simultaneously. The settings in the Positioning Method are used for rigid registration, while the settings in the Bundle Constraint are used for constraint adjustment.
Auto Split and Manual Split processing in the relative AT bundle phase utilizes a multi-machine parallel processing strategy, improving the computational efficiency when using cluster for large-scale AT. For cases with more than 8,000 photos, it is recommended to divide into sub-blocks.
Auto Split: Each photo has position information. Photos acquired along regular flight lines are automatically divided into sub-blocks based on the set number of photos per sub-block.
Manual Split: Used when some POS data is missing or unavailable, dividing photos into sub-blocks based on photo groups.
No Split: Recommended when the total photo count is less than 8,000.
The AT process involves moving from one stage to the next after completing the previous task. Multi-task stages can run in parallel on multiple machines, while single-task stages are processed on a single machine. (It is normal for some engines waiting during the AT process.)
Auto Split processing in the absolute AT bundle phase utilizes a multi-machine parallel processing strategy, improving the computational efficiency when using cluster for large-scale AT. For cases with more than 50,000 photos, it is recommended to divide into sub-blocks.
Auto Split: Significantly improves Absolute AT speed. If the number of control points is less than 10, it is not recommended to split block. The default setting of 50,000 images per block is recommended. Setting it too small may result in sub-blocks with no control points.
No Split: A single task is generated and can only be processed on one machine.
The AT process involves moving from one stage to the next after completing the previous task. Multi-task stages can run in parallel on multiple machines, while single-task stages are processed on a single machine. (It is normal for some engines waiting during the AT process.)
It is recommended to select Efficiency Optimize for oblique image data. Deselecting Efficiency Optimize requires more memory and longer computation time. A computer with 64GB of memory can process up to 100,000 images.
For oblique data, it is recommended to submit with the default parameters. And select the AT Optimization.
Recommended parameters for orthophoto scenes.
Image Paris Filter Level--Medium
Default parameters for AT, the results are distorted.
Recommended parameters for AT, the results are normal.
Search Range--Diameter of one circle of the flight
Image Paris Filter Level--Medium
Default parameters for AT, some images lost(blue area).
Only select Pose Metadata for AT, the number of lost photos is reduced.
Recommended parameters for AT, all images are in net.
Recommended parameters for 5-lens rule flight.
In the relative orientation stage, select Pose Metadata to improve AT absolute accuracy.
2.4.2 Resubmit after Relative Orientation
AT absolute accuracy can be improved by selecting Pose Metadata and applying Bundle Constraints through Custom parameter settings. The parameter settings are as follows:
◉ Custom
◉ No Split
Photo Selection--Photos in Net
Tie Points--Keep
Recommended parameters for close range scenes.
The left image shows the model result with default parameters, while the right image shows the model result after AT optimization.
For oblique data, it is recommended to submit with the default parameters. For multi-machine processing with block division, adjust Auto Split block image number.
Select Control Points and Pose Metadata to improve AT absolute accuracy.
Use only control points for rigid registration, without Bundle Constraint. Typically used for low-accuracy AT position adjustments.
In the project list, right-click on the processed block to open the AT optimization interface.
AT efficiency decreases, but relative accuracy can be improved, effectively enhancing model quality. The data processing flow is as follows: it can be selected directly during the relative orientation stage or applied based on the relative orientation results.
1. Select the AT Optimization settings when submitting the relative orientation.
2. AT Optimization settings for relative orientation results.
Case: If the scene contains a large number of urban buildings, it is recommended to select this option in the relative orientation stage. If structural issues are found in the modeling results, as shown in the screenshot below, AT can be optimized through secondary submission.
After merging multiple blocks with completed relative orientation, apply seam optimization to reduce the AT layering phenomenon between blocks. Parameter settings are divided into two cases.
1. Multiple AT blocks roughly aligned
When POS is input for relative orientation, it can be considered roughly aligned. If absolute orientation has been applied, it can also be considered roughly aligned.
2. Multiple AT blocks not aligned
When POS is not input for relative orientation and no absolute orientation has been applied.
Merge multiple blocks with completed absolute orientation, then perform this operation to reduce seam layering.
The data processing flow is as follows: Perform absolute orientation, merge AT blocks, and submit absolute orientation seam optimization.
AT setting is as follows:
Case: Considering the modeling scenario of multi-period data production, the whole survey area is divided into several blocks. After relative orientation respectively, it is a safe and reliable way to connect the edges of AT blocks through control points.
However, the collection of control points is affected by various factors, leading to situations where control points cannot be collected between AT blocks. In addition, some control points may be obstructed and unusable during the point marking process. Photos at the edges of the AT, with lower overlap and weaker constraints, have a higher probability of flying out. The above three factors will cause layering problems in the model when merging reconstruction, due to insufficient common control point constraints between blocks.
The image below shows the Get3D Mapper Data Manage interface, with different colors indicating different AT blocks.
The left image shows before optimization, and the right image shows after optimization.
This optimization targets small-scale layering issues that automated AT cannot resolve. By adding tie points in the layered area and marking as many visible photos as possible. If the RMS reprojection error exceeds 3 pixels or the reprojection error(relative) of some photos exceeds 5 pixels, the AT relative accuracy in that area is abnormal, leading to structural layering in the model. This operation helps assess AT relative accuracy. The processing flow is as follows:
Select Optimize by Tie Control Points, and the remaining parameters can be submitted with defaults.
After adding the tie points, use these points to optimize AT.
The left image shows before optimization, and the right image shows after optimization.
The software displays a warning about inconsistent camera intrinsic parameters when the parameters for a photo group do not match the parameters acquired by the software. There are two solutions:
Solution 1: Organize the photos by resolution.
Solution 2: Re-import the photos, selecting the Add Ground Photos.
In Get3D Mapper, all photos in a group should use the same set of camera intrinsics. When importing photos, group them by folder. During data collection, if different devices or focal lengths are used, or if automatic photo rotation occurs, make sure to manually place photos with different parameters in separate folders.
Significant shadow variations caused by lighting changes during aerial photography can lead to large amounts of photo loss in AT, especially when flights are captured in the morning and afternoon. If all photos in one flight are lost at the seam between two flights, you can copy the relative orientation AT results, retain only the vertical images, and check the seam areas between the two flights.
Case: Pink and bright green represent two different flights. In the following images, the tenth flight line from the bottom (highlighted in the red box) has no photos in the net.
The following image shows adjacent original photos with significant shadow variation, where the overlapping areas appear completely different on the photos.
Solution:
Office Work Solution: Perform separate AT processing for the upper and lower sections, then use control points to merge the two AT blocks.
Field Work Solution: Re-fly and re-collect data to resolve the issue.
Field Work Recommendations:
Photo loss in AT processing is caused by issues during data collection. When collecting data in the field, try to avoid collecting the data of neighboring flights separately in the morning and afternoon of the same day, especially on sunny days with large light difference.
