XangleCs documentation

(for Xangle users and anyone working with multi-camera photogrammetry / gaussian splat / 4D capture)

In most cases, RealityScan can register your camera positions directly from the original photo set. However, there are several situations where relying on automatic feature matching is not enough. In these cases, using a reference calibration shot dramatically improves accuracy and ensures consistent results across multiple datasets.

Below are the three most common scenarios where you must use a reference shot:

1. When your subject lacks distinct features

   Examples: smooth skin, white or all-black clothing, glossy or uniform materials.

   These conditions dramatically reduce the number of usable tie points.

2. When you need consistent calibration across multiple datasets

   This applies to any capture session where the rig must remain identical - e.g., multiple takes of the same pose or repeated scans over time.

3. When capturing for 4D sequences

   Even tiny inconsistencies in per-shot alignment will break temporal coherence. A locked calibration guarantees that all frames align to the same coordinate system.

What You Need

A simple calibration box (a cardboard or wooden box) with printed RealityScan markers attached on all visible faces.

Marker documentation: https://rshelp.capturingreality.com/en-US/tools/detectmarkers.htm

This box does not need to be fancy. The only requirement is that the markers are flat and clearly visible.

How to Capture Your Calibration Reference

1. Take one shot of the calibration box with all cameras

   Place the box near the subject position, ensuring markers face as many cameras as possible.

   (Example: 203-camera array)

2. Import images into RealityScan

   Drag and drop the full set into the application.

3. Detect the markers

   Open Alignment → Detect Markers.

   Make sure a reasonable number of markers are detected (e.g., “69 points found”).

4. Align the images

   Run Alignment → Align Images.

   The calibration should produce a clean, stable reconstruction with all cameras positioned correctly.

5. Export the calibration (XMP)

   Still under Alignment, choose Export Metadata (XMP).

   Set Camera Export Mode → Export as Locked.

   This generates one XMP file per camera, containing its position and orientation in 3D space.

How to Reuse Your XMP Calibration Files

1. Copy all XMP files from the reference shot into your new dataset folder.
2. Drag and drop the images and XMP files into RealityScan

   You will see a small pin icon next to each image—this confirms that the camera poses were successfully loaded.

3. Run Alignment again Use Alignment → Align Images. RealityScan will keep the camera positions locked and only compute the 3D structure.

Validation Tips (Important & Often Forgotten)

To ensure your calibration is solid:

• Check that all cameras are present in the reference alignment.

  Missing or unaligned cameras break the entire calibration.

• Ensure markers cover at least 4 sides of the box.

  A flat marker sheet is not enough. You need depth for proper triangulation.

• Keep the reference box at the actual subject location.

  Avoid placing it too far from where the actor/object will stand.

• Do not reuse calibration across different days unless the cameras physically cannot move.

Best Practices & Common Pitfalls

• If any camera moves, even 1 mm → redo the reference shot.

  This includes tripod tightening, cables pulling, or a colleague bumping the rig.

• Take multiple reference shots

  We recommend:

  • One at the beginning of the session
  • One at the end
  • Additional ones if someone (or a giraffe) kicks a camera
• Use consistent lighting for calibration.

  Harsh shadows or blown-out highlights on the markers reduce detection quality.

• Use high-quality prints of the markers.

  Low-resolution or slightly blurry prints reduce detection accuracy.

Why This Matters

Using a reference calibration:

• Ensures consistent multi-take sessions
• Prevents drifting camera positions in 4D reconstruction
• Improves stability with low-texture subjects
• Allows you to align many datasets into the same coordinate system
• Reduces failed alignments and manual cleanup

For any serious multi-camera or volumetric workflow, a reliable reference shot is the foundation of clean reconstruction.