Friday, June 3, 2016

Multi View stereo - Chickering monument

Set of 19 frames (1080x1920 pixels) for which we want to build a dense 3d reconstruction (point cloud).

We are gonna use Structure from Motion 10 (SfM10) to extract the camera poses (Structure from Motion) and Multi View Stereo 10 (MVS10) to reconstruct the dense 3D scene (Multi View Stereo).

Animated gif showing the dense 3d reconstruction built by MVS10 using a minimum number of image points (per 3D point) equal to 3. There are 2,559,264 3D points in the dense reconstruction.

Input "mvs10_input.txt" used:

duh.nvm
200
0.5
32
0.9
30.0
10.0
4
0
4
1
0.5
3
2.0
1

Let's see if we can get a more accurate 3D reconstruction by increasing the minimum number of image points (per 3D point) to 4.

Animated gif showing the dense 3d reconstruction built by MVS10 using a minimum number of image points (per 3D point) equal to 4. There are 1,439,609 3D points in the dense reconstruction.

duh.nvm
200
0.5
32
0.9
30.0
10.0
4
0
4
1
0.5
4
2.0
1

Clearly, you end up with a dense reconstruction that has fewer 3D points but is a bit more accurate.