Public Lab Research note


Photo Reconstruction from KAP at Acacia Reservation

by dbenjamin | September 04, 2015 14:07 | 69 views | 3 comments | #12199 | 69 views | 3 comments | #12199 04 Sep 14:07

Read more: publiclab.org/n/12199


What I want to do

We decided to check out Acacia Reservation, which was an old golf course that has been allowed to grow freely for the past year. We also wanted to test our new helikite using the Saturn V aerial rig.

My attempt and results

We used Mode 2 (9,9,8,6,3, slow) on a Canon S100 RGB Camera mounted on the Saturn V w/ Skyshield v2.041.

IMG_5019.JPG

IMG_5018.JPG

I discovered afterwards that the camera was only taking a photo every other angle, so we did not have enough photos to input into ICE, but I still tried to perform a 3D reconstruction using OpenDroneMap. I was unable to create a textured mesh, but still produced a decent dense point cloud:

http://plas.io/?s=https://cleveland-metroparks.github.io/acacia-point-cloud/acacia.las

You can download the original ply file here and view it using software such as Meshlab.

Here are a few photos from the flight:

IMG_1736.JPG IMG_1738.JPG

IMG_1735.JPG

IMG_1732.JPG

IMG_1737.JPG

IMG_1733.JPG

IMG_1734.JPG

Questions and next steps

Why did the camera only take photos every other angle? Is it a CHDK setting, or something in SkyShield?

We would like to fly acacia a few more times to get a full reconstruction. It would be interesting to see how the land has changed since the Metroparks acquired the property.


3 Comments

What an eerie, apocalyptic place. This is what happens to golf courses when maintenance ceases. I hope someone is documenting the succession.

I have not experienced the "missing every other shot" with the same camera and SkyShield version you have. The first thing I would check is the CHDK remote settings. They should be something like:
Enable Remote [] (selected)
Switch Type [OnePush]
Control Mode [Quick]
Enable Synch [ ] (not selected)

When Synch is enabled, the camera shoots on every other pulse, so that is suspect.

The mesh looks good in Meshlab. Did you sort the photos that went into the mesh so only the ones aimed downward were used? Did you also remove near duplicates? I'm just curious because I wonder what the best protocol is for capturing photos sets most appropriate for structure from motion. The SkyShield needs a mode for sfm to capture just the downward angles. But I don't know what angles are best for sfm.

How much helium does the helikite require? Any problem lifting the rig?

Cool result despite the missed photos.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


We have some photos from May that I'm hoping to compare to. The difference from the ground is just astounding.

That was exactly the problem, thank you! I'll be sure to double check that option next flight.

I selected two sets of rotations but removed the horizon-facing photos. Even if there were some near-duplicates, it would do nothing or maybe improve the process. Given the paucity of this dataset, I didn't bother checking. I think for structure from motion, 60-75% overlap is ideal. I'm not sure how to translate that to angles on the SkyShield, because it would at least somewhat depend on the altitude.

We have the Skyhook 3.0, which requires 3 m3 helium.

Reply to this comment...


Nice work. I was able to download your OpenDroneMappoint cloud ply, view it in Meshlab. Cropped snapshot from model is interesting. snapshot00_crpd.jpg

Reply to this comment...


Login to comment.

Public Lab is open for anyone and will always be free. By signing up you'll join a diverse group of community researchers and tap into a lot of grassroots expertise.

Sign up