Creating Ortho/DEM from large amounts of UAS imagery WORKFLOW help

[Gator_UAV]

Hello all,

First of all, thank you for all the help on these forums so far. I'm a beginner in Agisoft and Photogrammetry and am learning quite a lot from here.

My question is about the workflow for my project. I have about 3,500 photos of an area that is ~1.2 square miles. The camera has pretty accurate position data recorded along with omega, phi, kappa, and velocity. I also have GCPs for the whole area. I'm using Photoscan Pro v1.0.4 Build 1874.

I've read the tutorials Agisoft has provided, and the manual, but I'm having trouble figuring out the best workflow for my situation, specifically involving chunks.

For instance, right now, my workflow looks like this:

1. Chunk up cameras. I use ArcGIS to show all our cameras, merge them (because the are from multiple flights),  and then select relatively arbitrary rectangular chunks that overlap with ~2 photos from each line. So that the chunks look like this:

2. Add these photos into the project in their respective chunks and import their locations. Turned out to be ~12 for this project specifically.

3. Align the photos.

4. Build mesh

5. Manually place and refine GCP markers, and import their locations.

6. Here is where I get a bit confused, especially in regards to combining the chunks. I'm not sure on the process after this, when the chunks should be combined, and with what technique. Should it be done before the GCP's are marked? Is this workflow fundamentally flawed because the project should be chunked after the GCPs are marked? When should the cameras be optimized?

Any information at all would be greatly appreciated! Thank you all for your time!

[andyroo]

Hi Gator_UAV,

Based on my experience with camera calibration (thanks to Porly's advice), I would recommend doing a "precalibration" project for a smaller area complete with entering GCPs and optimizing alignment. Then save your camera calibration and add it to the chunk or chunks for your larger project. Precalibration takes out a lot of the warping ("smiles" or "frowns") that results from drift at the edge of the image due to spherical aberration or whatever effect it is (I forgot my photogrammetry class lingo). This could improve your chunk merging.

Also if you have a machine that allows it, you could do the alignment and GCP step first (should need about 50GB memory), then optimize alignment, then chunk up the model, build dense cloud, and mesh. That saves you the time of placing duplicate GCPs for the overlapping area, and ensures better alignment for the chunks since they were all aligned together. There are good memory guidelines here.

Also, FYI I've started to generate models for my entire area and just output the region of interest for orthos and DEMs. I was doing 4 chunks with 40 million polys each and just upped it to one chunk with 200 million polys. Again, the memory guidelines should tell you what you can do. I'm lucky enough to have 192GB RAM, but you should be able to do that with 80GB in High quality mode.

You may find some of my detailed description of what I do in this post on my "large project" workflow.

Finally, I would recommend trying high quality before you try ultra - at least for my camera/platform I get better DEM reconstruction in HQ mode than in ultra, especially for areas with trees. And the time/memory requirements are much less. Although I don't see any of those in your sample area...

Andy