Parts of point cloud not showing up on textured model

Hi All, hope everyone’s keeping well in these strange times.

I’ve been playing around with 3D modeling for the past few months and eventually came to WebODM as a favorite tool (open source whats not to love, easy to use and great UI) My use case is 3D modeling telecoms towers/other infrastructure but no matter what i cant seem to get decent models with all elements intact (antennas in particular seem to be disappearing from the model)

The attached are a series of photos I took with a phantom 4 pro and console output from WebODM. This is a model I made in which the point cloud clearly has a visible tower but when I switch on the texture it competently ignores it. Any help in figuring this out would be greatly appreciated.

Drop box link to photos and console:

1 Like

Sorry couldn’t attach these to the post.

image

image

1 Like

It looks like there aren’t enough points at the base of the tower, which probably resulted in a “floating” tower, so WebODM cleaned up by mistake thinking it was noise.

This is due to the “removeIslands” flag that is always enabled (https://github.com/OpenDroneMap/ODM/blob/ef174de8af8dbb87379fa882f20ede34c751aef1/opendm/mesh.py#L129)

There’s no way to turn that flag off for the time being (unless you patch it), but it highlights an important fact: you don’t have sufficient information at the base of the tower.

Perhaps if you set mve-confidence to zero you will get enough points to keep it, but note it will also increase noise in the overall reconstruction.

3 Likes

Thanks for the reply. I see that most likely explains it. I could possibly try but it seems like an important enough flag to keep intact, possibly I could try increase/decrease the sensitivity at the users request. I’ve actually gone back out today to take more photos with a better mix of parameters height, distance from tower and gimbal angle so hopefully that would fix the issue, I’ll run it again overnight (large data set this time 400+ images) and hope for the best.

Just out of curiosity, would webOdm care if I included just pictures of the base, say do a few rotations at a much lower altitude around just the base, could the feature detection match these from the photos taken above the tower then?

2 Likes

Absolutely. You can also use different cameras (say, a hand-held one).

1 Like

If you re-shot just the base material and combined, this often works, especially if the lighting and vegetation is similar to the previous flights. The algorithms are pretty robust to lighting changes, but in general it’s better to shoot at the same time of day, similar cloud cover for optimimum matching (this could just be my superstition though, so take with a grain of salt).

1 Like

Excellent, I will try with the new images and if not I’ll combine all the data sets I have plus take a few new ones of just the base. :slight_smile: thanks for the assist …

2 Likes

Best way to find out would be a little experiment, I’m going to re shoot just the base tomorrow, depending on the results of tonight’s outputs. If I need to redo I’ll take your advice onboard and head out at the same time. Thanks for the reply.

2 Likes

I just happened on this link recently and thought it may give some help on your data gathering.
https://support.pix4d.com/hc/en-us/articles/202557459-Step-1-Before-Starting-a-Project-1-Designing-the-Image-Acquisition-Plan-a-Selecting-the-Image-Acquisition-Plan-Type#label11

1 Like

Hi Julian, thanks for the link very informative. I’ve been scouring the internet for stuff related to 3D models of vertical structures, towers, poles etc I’ll compile a tutorial when I eventually get it working properly and post a link :slight_smile:

3 Likes