Large scale 3D reconstruction

Written by Paul Bourke
June 2021

Project 1: June 2021

The follow documents a large scale, both spatially and with respect to data, 3D reconstruction exercise.
Input: 10,000 drone photographs (20MPixel), 8.5km, 20 GCP points.
Output: 50 million triangles, 4 x 16K textures.

Final textured mesh showing GPC marker locations.

Sparce point cloud of a small section

Untextured mesh of a small section

Mesh confidence, highest along the rail tracks where the drone images are centered.

Sample renders of the 3D model.

Project 2: September 2021

Input: 32,000 photographs from 3 drones. 10km. 16 GCP points.
Ouput: model split into 8 pieces (shared edges), each piece approximately 10 million triangles and each with 2 x 16K textures.

Confidence data

Screen captures of the 3D model.