In this article:
- Infrastructure (One-Time Setup)
- Data Validation (Before AT)
- Aerotriangulation
- Reconstruction
- Outputs
1. Overview
This checklist combines verification steps and practical guidance to help ensure optimal PhotoMesh performance and high-quality results throughout the project lifecycle.
2. Infrastructure (One-Time Setup)
Hardware and Software
- Please see the latest Photomesh and Fuser Hardware and software requirements: https://support.skylinesoft.com/hc/en-us/articles/360012991599-Software-and-Hardware-Requirements
- Working folder on SSD for fast read/write
- Fast network for optimizing fuser I/O
- Up-to-date graphics drivers
- No pending Windows updates (Some PhotoMesh processes may break when updates are pending)
Permissions
- Give Full Control permissions to the PhotoMesh Windows user on project and temp folders
- Run PhotoMesh & fusers from Administrator account if possible (Ensure Full Control permissions for the Fuser and the PhotoMesh Manager).
- Open the PM project from the UNC path for AWS/Pool users.
Multi-Fuser Environments
- Add fusers proportional to available RAM & CPU cores (E.g., a PC with 64 GB RAM and 16 logical cores can run 4 fusers simultaneously)
3. Data Validation (Before AT)
Camera and Orientation
- Verify correct camera properties for all collections (Note that nadir and oblique collections use different properties).
- Verify correct orientation: Project neighboring photos “on terrain” to verify photo position and orientation accuracy.
Photo Coverage and Overlap
- Ensure no holes or missing image rows
- Verify proper photo overlap:
- Forward overlap (in the direction of flight): 75% to 90% (80% recommended)
- Side overlap (between neighboring rows): 70% to 90% (80% recommended)
Photo Quality
- Remove corrupted or duplicate frames: Check photo folders for duplications. Delete takeoff/landing photos common in drone data.
- Check for consistent lighting and minimal shadows across collections. If needed, run color correction presets such as CLAHE, RGB histogram stretch or use third-party tools.
Coordinate System
- Verify correct coordinate system settings (Geoid/Ellipsoid/Relative to Terrain/Absolute heights). Incorrect definitions may place photos underground or in incorrect locations: Verify that photo icons are visible and correctly positioned. Project photos on terrain and inspect their frustums to confirm correct height reference, orientation, and alignment with the terrain.
4. Aerotriangulation
- Display the different Coverage Maps to assess coverage and overlap in the defined AT area, and adjust the AT area polygon as needed.
- Select appropriate AT settings for your data (e.g., external inputs like INPHO, Bingo or Stellacore already include calculated camera positions and EO, so lock the current values and set the Calculation Mode to Do Not Calculate AT (Fully Trusted) or Refine AT.
- Run an AT Only build to ensure a precise AT before running the full PhotoMesh project.
- After the initial AT is finished, inspect the calculated photos and the sparse point cloud. Look for anomalies, holes in the sparse cloud, or large deviations between the original and calculated photo positions.
- Review the Quality Report and AT log for information about errors and missing information, and to identify photos with high mean median errors.
- Add GCP and tie points (It is easier to add control points once after the initial AT solution once external orientation information is available).
- Set an appropriate Target Photos per AT Tile, ensuring the number of tiles suits your project and your GCP scattering (Minimum of 4 GCPs per AT tile).
- Check tie/check points: Review the points with the highest errors and resample if necessary.
- Iterate as necessary: Adjust AT parameters, collection properties, or control points, and rerun AT until AT results are satisfactory.
5. Reconstruction
- After aerotriangulation, build a few selected tiles (Point Cloud → Texture) before running a full build. If the overall quality meets expectations, proceed with the full build.
- Choose optimal Build Parameters: In the Reconstruction tab, adjust parameters like Point Cloud Quality and Min. Number of Matches to balance between processing time and output detail. (For example, setting Point Cloud Quality to Optimize can speed up processing while still providing good results).
- Select the Tile Split Method – Fixed 2D grid or Adaptive 3D boxes: Fixed 2D grid divides the project area into uniformly sized tiles on a horizontal plane. Adaptive 3D boxes divide the project, along the horizontal, vertical and depth axes, into 3D tiles based on the memory requirements of each section of the reconstruction area.
- For Amazon Web Services (AWS) users, ensure you are using the correct machine type: For point cloud and model steps, select machine types optimized for reconstruction, such as M6i.2xlarge, M5.2xlarge, or M5dn.2xlarge on AWS. This ensures efficient processing and high-quality results. (PM processes that use CPU: AT, Point cloud, model, output. PM processes that use GPU: Photo preparation, Texture, Ortho output)
- Review and retouch mesh: After reconstruction, use manual retouch tools to fix mesh imperfections like bumps, irregular surfaces, or floating artifacts. You can flatten polygons or fill areas with a single color for a cleaner output.
- Rebuild selected tiles if errors are found: Do not leave unfinished reconstruction tiles since this causes holes in the final output. View reconstruction tiles to verify that all tiles are green, indicating that all steps through project output are complete.
- Create a new build version with different parameters for further refinement: New Build Version – Copy AT. Test alternative reconstruction settings.
6. Outputs
- Generate expected outputs in a small reconstruction area:
- 3D Mesh
- Orthophoto
- Point Cloud
- DSM/DTM
- Validate final outputs visually in the target applications: Cesium Sand Castle, QGIS, etc.
- Check for possible issues like holes in the mesh, missing tiles, pixel size holes in Orthophoto, XYZ shifts, missing LOD.
- Verify output heights against known references: Use GCPs or known points for reference. Note that third-party apps may use custom terrain and reprojection methods.