Photogrammetry Results from the Jemez Historical Site, Jemez New Mexico

Data Sets

      Below are the three photo sets that were collected on January 20th, 2018. Many thanks to the New Mexico Park Service for permission to fly the drone for the first two sets. A quick explanation, the grid set does a better job of mapping the entire area, while the circular mission does a better job of mapping objects inside the circle, and uses fewer photos. The ground photos were taken using a much better camera, but cannot see the top of walls, and results in a very large data set. Combining data sets for even better results is a topic for the future.
Jemez_Grid3ScanPoints_500 Grid mission over Jemez Cathedral and surrounding area
GPS data included

January 20th, 2018, 12:16pm-12:31pm

Download (509MB)
Jemez_CircleScanPoints_500 Circular mission over Jemez Cathedral
GPS data included

January 20th, 2018, 1:29pm-1:33pm

Download (200MB)
JemezInside_1(Pix4D)_500 Incomplete ground photography inside Jemez Cathedral
No GPS data

January 20th, 2018, 2:20pm-2:51pm

Download (2,483MB)

Models

      There are two photogrammetry paths represented here.

      First, Pix4D is a piece of commercial software and is very easy to use. It requires little more than create the project, select the folder with the pictures, and push go. The second stage can use a large amount of RAM.

      Second is a combination of COLMAP for image feature extraction, matching, and sparse reconstruction; followed by OpenMVS for dense reconstruction, meshing, and texturing. The biggest challenge with this pipeline is that while both COLMAP and OpenMVS support the fisheye lens on my drone, the export/import path between them does not. I currently run COLMAP to completion using the SIMPLE_RADIAL_FISHEYE lens model and spatial matching since I have GPS data, export "undistorted images" with a min_scale setting of 0.8, then close and reopen COLMAP. I then create a new project and use the undistorted images as the input and the SIMPLE_RADIAL lens model, run COLMAP a second time, and export the *.nvm file. This seems a bit silly, so I am looking for a better way.
      OpenMVS is fairly straight forward. Import in the *.nvm, then Densify the point cloud, mesh the densified cloud, then texture the mesh. This can use a huge amount of RAM.

      The *.obj files are just as the software left them. The Blender files have the model straightened with Z-up and a infinite light source (sun) set up approximately as the sun was when the photos were taken.

      Now, as fow what model to use, the best looking model of the main cathedral seems to be the circular mission processed by COLMAP and OpenMVS. However, everything else is sub-par due to poor camera coverage. Also, both of the COLMAP-OpenMVS models are very large and may overload normal computers. For general usage, the grid mission processed by Pix4D would be my first choice.

Jamez_Circle(Pix4D_CloseCrop)_500 Circular Mission converted using Pix4D, Cathedral only Download *.obj (10MB)
Download *.blend (18MB)

*.stl for 3D printing
Download *.stl (7MB)

Jamez_Circle(Pix4D_WholePark) Circular Mission converted using Pix4D Download *.obj (27MB)
Download *.blend (98MB)
Jemez_Circle(COLMAP-OpenMVS_SRF_spatial_UD(0.8))_500 Circular Mission converted using COLMAP and OpenMVS

This is the highest quality main building, but everything else suffers.

Download *.obj (99MB)
Download *.blend (374MB)
Jemez_Grid3(Pix4D_WholePark)_500 Third Grid Mission converted using Pix4D

This is my first choice for general usage.

Download *.obj (31MB)
Download *.blend (114MB)
Jemez_Grid3(COLMAP-OpenMVS_SRF_spatial_UD(0.8))_500 Third Grid Mission converted using COLMAP and OpenMVS Download *.obj (180MB)
Download *.blend (684MB)
JemezInside_1(Pix4D)_500 Model Attempt from the ground based photo set.

The computer with Pix4D installed will need more RAM to complete this task. That may take awhile. Shown is the point cloud that Pix4D creates at the end of the sparse point cloud generation step. It looks very promising.

none

Video

Flyby A "quick" raytraced video of the model. (only took 26 hours for 10 seconds of video). I had tried combining the circle and grid datasets together for this model. Download *.mp4 (8MB)

Main Page