Update 19 Aerotriangulation (AT) settings

 

The purpose of this document is to help users select the correct AT parameters depending on their scene and acquisition configuration, as there has been major changes integrated into the AT menu, the product team have tried to explain the procedure as detailed and exhaustive as possible, but before reviewing this document we remind the user that:

• If the basic acquisition rules have been followed, the user will get positive AT results without making changes to the default settings, thanks to a robust and constantly evolving AT algorithm. For more details on the Acquisition rules, an Acquisition Guide and Drone Capture Guide are available in the ContextCapture installation folder.
• The camera and AT parameters are automatically set depending on the imported photos and metadata, as this is the case a first AT with the default settings is not unfavourable.
• Thanks to ContextCapture’s block structure architecture and panel, it is always possible to test several AT configurations and still have the ability to revert back to a previous result

 

Contents

1       Components (optional).

2       Positioning, georeferencing.

1.1         General rules.

1.2         When to use Adjustment constraints.

1.3         When to use Final rigid registration.

1.4         Georeferencing using control points.

1.5         Georeferencing using photo metadata.

1.6         Georeferencing using laser scan point cloud.

1.7         Scaling using user tie-points and positioning constraints.

1.8         Use Targets (QR codes, Apriltags or ChiliTags).

3       AT settings.

3.1         Poses and tie points.

3.2         Optical parameters.

3.3         Other options.

3.4         User presets (advanced usage).

 

 

 

1      Components (optional)

This option is only available if a partially complete AT (for which some photos remain not consolidated) has been completed on the photo block. For a partially complete AT, the 2 following settings are available:

• Use all photos. Select this option if you want to have all photos calibrated and added to the main component
• Use only photos of the main components. Select this option if you consider that the photos not in the main component are not relevant for your result / model.

Note: You can have an overview of the non-calibrated photos by doing the following:

• display photo not in the main component in 3D view filter
• list of photos not in the main component in the Photos tab
• percentage and number of calibrated photos in quality report

Figure 1: option to display only non-calibrated (without component) photos in the 3D view

2      Positioning, georeferencing

1.1     General rules

• Any metadata with a high level of accuracy (precise GPS: 5cm or less) can be selected as adjustment constraint
• For metadata with a lower level of accuracy (floating DGPS or raw GPS: 10cm or above), this can only be used for Final rigid registration
• If only 1 georeferencing source is available and can be used as an adjustment constraint (such as the photo positioning metadata in the above screenshot) it will be used for the final scene registration in the AT backend, as this is the case it is not necessary to check the Final rigid registration

1.2     When to use Adjustment constraints

• When selecting the adjustment constraint setting the metadata/GCPs will be used for the AT intermediate stage of photo-matching. It therefore has the power to change the photos connections (tie-points) and contort the photo block.
• Any data which has a high level of accuracy (5cm or less) can utilise the adjustment constraint setting.
• Several constraints can be chosen, but the user must make sure that these constraints are consistent (no Z offset between photo metadata and GCPs for instance) or it may cause the AT to fail.

1.3     When to use Final rigid registration

• When selecting the Rigid registration setting the metadata/GCPs  will be used in the final AT step: georeferencing/scaling. It will only be used to ‘block-shift’ (rigid translation, rotation, scale factor) the photo-block to the defined location but does not have the ability to contort the block or modify the connections between photos
• Any georeferencing data can be used as a rigid constraint. Even a single control point can be utilised to georeference the scene.
• Only one data source can be used as a Rigid registration constraint.

1.4     Georeferencing using control points

If control points are utilised in the block these will be set as an adjustment constraint by default, as they normally have a high level of accuracy.

• In case the user created some approximate control points (doubtful data source, Google earth marker…), it is recommended to set these as a rigid constraint instead.

1.5     Georeferencing using photo metadata

If the images have pose metadata (geotags or imported pose data), ContextCapture set this as a rigid constraint by default, as most drones still have a ‘raw’ GPS.

• If a RTK set was used, or if the imported pose metadata comes with a high level of accuracy, it is recommended to change the settings to an adjustment constraint.

1.6     Georeferencing using laser scan point cloud

• Use as an adjustment constraint: If the user has imported accurate LiDAR point clouds and wants to contort / modify the photoblock to make it align more accurately with the point clouds data.
• Use for final rigid registration: If the point cloud is noisy and with a low resolution this is the best option to make sure there photo block is not incorrectly contorted/modified.

1.7     Scaling using user tie-points and positioning constraints

This setting can only be used for the final Rigid Registration and will only scale the data (no georeferencing)

 

• Use this option to scale the photoblock if no georeferenced data source is available

1.8     Targets (QR codes, Apriltags or ChiliTags)

• Use this option if you have printed and placed several targets over your scene during the acquisition. Note that selecting this option automatically enables the Target extraction option in the AT settings

 

3      AT settings

The following parameters have an incidence on how the photos will be automatically connected together using Automatic tie-points

3.1     Poses and tie points

Four main parameters are available plus the Advanced menu

3.1.1        Compute

The tie-points and photo poses are (re-)calculated.

• Use this option when running a first AT with non-georeferenced images (in this case the Compute option is the only one available).
• Use this option when your current AT is unsatisfactory and when you have added one or several constraints (User tie-point, Block type and viewing distance, different pair selection mode) in order to improve the AT result.

3.1.2        Adjust

The tie-points are (re)-computed but the image pose will be adjusted from the current one.

You can use this option when:

• Running a first AT with accurate poses (position computed with a high-end GPS receiver, e. 1 - 10 cm relative accuracy)
• Your current AT result is satisfactory, but you still want to improve the accuracy after adding additional surveys.

3.1.3        Extend

The engine is using the current main component, or for a first AT, the biggest subset that it manages to consolidate, and is gradually adding disconnected subsets of images to connect it to the main component.

Note: this option was called ‘Complete incremental engine’ in the previous versions

You can use this option:

• When merging 2 adjacent block which have an altitude offset because of changing GPS elevation settings.
• When you update a dataset by adding additional photos to an existing block.

3.1.4        Lock

The engine will keep the image pose and tie-points, and move straight to the final stage of the AT: scaling/georeferencing, this will accelerate the time taken to complete an AT.

You can use this option when:

• You have a complete AT but you want to run the final color equalization.
• You have a consistent 3D scene and you want to improve the georeferencing after adding additional control points.

3.1.5        Advanced poses and tie-points options

The Advanced… button allows you to fine-tune the Poses and tie-points policies, it is recommended to modify these parameters only after you have tested the default AT settings.

The user can manually modify the images Position / Rotation and the tie-point policy.

The user can also manually modify the following parameters:

• The Keypoints density: from Normal to High
• The Pair selection mode:
  • Default: the default pair-matching calculation normally used for the AT
  • Similar images only: it estimates the relevant photo to match based on photos keypoints similarity. It can be used on a building façade acquisition with non-repetitive details (old buildings). It can be useful when the set of photos is not sorted properly (e.g. photos that are next to each other on the list don’t overlap).
  • Exhaustive: Tests all photo pairing possibilities. Only use this for small datasets or airborne datasets with the rig mode enabled and block type: structured aerial dataset with minimum and maximum viewing distances
  • Sequence (one photogroup only): Follows the photo import sorting to match them
  • Loop (one photogroup only): Same as sequence but will also try to match the first photos with the last ones.

3.2     Optical parameters

In this section, the user will define the policy for the photogroups optical parameters: focal length, principal point, distortions…

3.2.1        Adjust main parameters

This is the default mode for all ATs. It will adjust all optical parameters (except ratio and skew) during the photos matching process.

• Keep this default setting unless you are applying your own camera calibration

3.2.2        Lock all parameters

The optical parameters values remain locked during the AT.

• Useful when a pre-established camera calibration is imported for the photogroup.

3.2.3        Advanced optical parameters

The Advanced… button allow you to fine-tune the Optical parameters settings. It is recommended to modify these parameters only once you have tested the default AT settings and requires a strong knowledge of the camera specifications.

3.2.4        Enable pre-calibration stage

This setting allows the AT engine to adjust the optical parameters in a broader range.

• Use this setting if you believe the default AT is not providing the correct camera calibration.

Note: this option is automatically enabled for a first AT without pre-defined optical parameters.

 

3.3     Other options

3.3.1        Target extraction

If targets have been placed on the scene during the acquisition, the corresponding target setting should be selected in this menu.

3.3.2        Blockwise color equalization

This option (enabled by default) is equalizing the images RGB information based on the AT correlation.

To have it applied in the 3D productions texture, you will have to leave the Color equalization mode on Blockwise, when selecting the Reconstruction Processing Settings.

3.3.3        Create Splats

SPLATS are small square elements generated for each tie-point at the last stage of the AT, to give a more intuitive overview of the 3D scene. This option is always enabled by default, but the user can  disable this setting if they considers that they will be able to analyze the 3D view with the 3D cloud of tie-points only.

• The splats mode is even more intuitive for large scale infrastructure and urban acquisitions.
• The splats can be regenerated at any time from the AT block General

3.4     User presets (advanced usage)

If a user wants to keep their own defined AT settings for later projects, they can save these in a user config file by using the Save button

There are other presets available in the ContextCapture installation folder. Press the  button to retrieve one of these presets.

Below are a bit of details on theses presets

3.4.1        Alternate engine (deprecated):

This robust engine is useful to get more photos consolidated on challenging datasets that is not possible the legacy AT engine (update 17 and previous), but it is not scalable. The current AT engine has proven to be both scalable and robust and as so this preset has a marginal interest now.

• Can be tested on small (a few hundred images) scenes with unsatisfactory results when using the default settings.

3.4.2        Do not check rigid registration reliability

This preset skips the scene consistency check based on GPS tags (i.e for inaccurate GPS on small scene). It can be useful for:

• Open trench acquisition using a smartphone (Many images acquired in a small space)
• Small smartphone / handheld camera acquisition with bad GPS connection (close to a tall building…)

3.4.3        Keep all components for adjustment

Us this preset to keep two or more subsets in the AT results, that can’t be physically connected together. Important! The poses policy needs to be on KEEP or ADJUST WITHIN TOLERANCE.

• Datasets with a large river in the middle etc.

3.4.4        Legacy engine (deprecated)

This preset uses the former AT engine (update 17 and previous). This is for users who want to test their dataset with the former engine.