Documentation for 3D Objects

#1

Good morning,

I’m currently taking courses, working towards the end of an Unmanned Aerial Degree and one of the assignments is:

  • Using WebODM/OpenDroneMap and ANY data you like (anything you used for this or previous GIS classes, anything collected yourself, or anything found online) to create both 1: AN ORTHOMOSAIC 2: A 3D MODEL.I recommend sticking to data sets under 75 images files, if you can help it. More is fine if you have a powerful PC.

I’ve not been able to find documentation on how to run just a 3D model. I’ve uploaded some images that I’ve processed in Metashape and Pix4D and they’ve failed miserably.

Can you point me in the right direction on how this can be done?

Thanks!

0 Likes

#2

A fellow UAS student!

Without knowing more about what your prof wants to know for your assignment, I would assume that he is talking about creating an orthomosaic and 3D model from the default options in WebODM. It’s the first thing you do after selecting images and naming the task.

If you are struggling because you have poor imagery, consider looking at the datasets section of this site and using one of the many over there. Alternatively, do a circular flight around some cars in a parking lot with a slewed camera angle. That would put you pretty close to the 75 images you are looking for. Try to avoid trees as ODM can struggle with figuring out where the branches are in 3D space.

0 Likes

#3

Hi Ryan,

The 3D render from my mapping images is what I ended up turning in. I think he may had a misconception of what 3D capability ODM had. We’d just finished 3 weeks working with Metashape and 2 weeks with Pix4D.

Not knowing if he knew something that wasn’t immediately documented, I tried processing a dataset that I made of an urn. It didn’t turn out so great. Unlike my mapping datasets, which I think are on par with products like DroneDeploy.

0 Likes

#4

ODM can absolutely create 3D models, and much like Pix4D and other photogrammetry packages, the difficulty is getting enough overlap between your images in order to get a good 3D model (in some ways, drone photogrammetry is much easier).

Tips:

  • Take your photos outside preferably on a cloudy day or in the shade so the light is relatively even but bright.
  • Make sure all your photos are in focus and have good depth of field. Here, the brightness helps, as then you can close down your aperture
  • Good 3D models need 83% overlap or so. That’s a lot of overlap. Here having a small thing you’re trying to reconstruct helps
  • Don’t use a turntable. You want your images to have the same lighting throughout because it helps with matching
  • Use surfaces that are not specular (not shiny) and have irregular patterns on them (not one even color.
  • This is a decent augment to the above tips: http://thehightechhobbyist.com/photogrammetry-tips-and-tricks/
0 Likes

#5

Thanks for the info. I had tried to process the dataset used to make this in Metashape, but it was less than successful.

0 Likes

#6

Can you be more specific?

0 Likes

#7

I’m not sure what went wrong with the Urn model I tried to process last time, I processed this tonight and it turned out fairly good.
WebODM 3D Schoolhouse Model

I’m curious if there’s a way to clean up, or remove points that aren’t wanted?

For reference, the same dataset processed in Metashape.

I’m running the Urn dataset again to see how it comes out.

1 Like

#8

Here’s the 2nd attempt to render the Urn dataset. WebODM doesn’t like it. This is the same dataset used to make the Urn model above on Sketchfab.

0 Likes

#9

This may be the problem with the Urn dataset. I did take all the photos with the urn on a turntable, keeping the lighting the same in every photo. Since it’s wood, I figured the surface texture would have enough variation to help find matches, which it did in Metashape, but I did a lot of masking and point cloud cleanup for that model. I’m pretty new to this, so I ahve a lot to learn still.

1 Like