Skip to work
RN
RN
Gaussian centres from the Calton Hill reconstruction, shown as a point representation.
[Gaussian centres from the Calton Hill reconstruction, shown as a point representation.]

A rendered view of the same early reconstruction, with visible gaps and artefacts. An improved version of the reconstruction pipeline is currently being tested.

Gaussian-splat rendering of the National Monument on Calton Hill.
[Gaussian-splat rendering of the National Monument on Calton Hill. Both Calton Hill images are derived from “The National Monument on Calton Hill (Source)” by Learning, Teaching and Web Services, University of Edinburgh. Reconstruction derivatives: CC BY-SA 3.0.]

The goal is a reconstruction that can be explored through space and time. Camera alignment, scene structure and moving objects contribute to the same replay.

BRAINDANCE selected workflow overview.
[From recorded footage to a navigable scene. Work in progress.]

The pipeline was also tested in a digital environment using multiple virtual cameras. Reconstructed movement was compared with the known motion in the scene.

Recovered position and reconstruction error in a synthetic five-camera bounce test.
[Synthetic multi-camera reconstruction test]