Aligning DSLR photos to SLAM LiDAR component without deforming internal camera poses

Development Asset Creation Photogrammetry
1

Hi,

I’m working on a 3DGS (3D Gaussian Splatting) scene and I’m stuck on a specific alignment problem. I followed the SLAM support tutorial ( SLAM Support | Tutorial ) using the local/Euclidean coordinate system section.

**My setup:**

- XGRIDS Lixel K1 SLAM scanner produces LAS point cloud + COLMAP trajectory (cameras.txt, images.txt, points3D.txt)

- Sony A7R IV (61MP DSLR) high-resolution photos of the same scene

- RealityScan 2.1

**What I’ve done (following the SLAM tutorial):**

1. Imported the K1 COLMAP data (Component 0) + LAS point cloud as a LiDAR scan. Registration: Exact, Georeferenced: Yes, Local:1 Euclidean. This creates the Laserscan component, which is my geometric reference at real-world scale.

2. The K1 Laserscan component is my ground truth — correct scale, correct geometry.

**My problem:**

I need the Sony A7R IV photos to be aligned onto the K1 Laserscan component’s coordinate system, but WITHOUT the K1 data deforming the internal consistency of the A7R IV camera poses.

When I align A7R IV photos alone, I get clean, consistent poses — good 3DGS training result.

When I align A7R IV + K1 together in one pass, the K1 LSPs pull the A7R IV poses — inconsistent internal geometry — bad 3DGS result (floaters, artifacts).

What I need is a **rigid body transformation** (translation + rotation + scale) that places the A7R IV component into the K1 reference frame, without any bundle adjustment modifying the internal relative poses of the A7R IV cameras.

**What I’ve already tried:**

- Align F6 with LSPs Absolute pose: Locked + A7R IV Absolute pose: Unknown — still deforms A7R IV internal poses

- Merge Components tool (after aligning A7R IV separately) — fails, creates duplicate components without merging

- Merge georeferenced components setting — not applicable, A7R IV component is not georeferenced

- “Merge components only” setting — doesn’t seem to exist in RealityScan 2.1

**My question:**

Is there a way in RealityScan 2.1 to align two components with a rigid transformation only — bringing the A7R IV component into the K1 reference frame while preserving the internal relative poses of the A7R IV cameras exactly as they were when aligned alone?

I’ve also considered the XMP “Export as exact” workflow (export XMP from the A7R IV-only project, then import photos + XMP into the K1 project and align) — would this preserve the internal A7R IV geometry and only solve for the global rigid transformation?

Any guidance would be greatly appreciated.

Thank you :slight_smile:

2

Hello @Stoug83

For such case you can use the option Lock pose for continue, which should keep the relative poses of aligned cameras. Select all aligned cameras form A7R IV and set the option to Yes. After that align the data with laser scans.

image

XMPs with such settings should work, too.

3

Hi Ondrej,

Thanks a lot for your answer! I tried both methods but got the same result each time.

Method 1 - Lock pose for continue on A7R IV cameras, then Align F6:

Component 0 (K1 COLMAP images): Enable alignment set to Disabled

Laserscan component (LSPs): Relative pose Exact, Absolute pose Locked, Enable alignment Enabled

A7R IV component (imported as .rsalign from a separate project where it was aligned alone): Lock pose for continue Yes, Absolute pose Unknown, Enable alignment Enabled

Result: Align F6 created two new components. One with the LSPs aligned with the Laserscan component, and one with the A7R IV photos in a completely separate coordinate system.

Method 2 - XMP Export as exact, then import photos with XMP into the K1 project:

Component 0 disabled, Laserscan component active with LSPs locked.

A7R IV photos imported directly with XMP files in the same folder.

Confirmed that Relative pose was read as Exact from the XMP files before launching Align.

Result: Same. Two separate components, A7R IV not aligned with K1.

Is there a step we are missing?

Thank you

4

Firstly align, then set Lock pose for continue for the component.

Also, as you are getting two components after alignment, you will need to use control points to merge the components together.