Building clouds for specific structures

[cadm8]

Hello all, I would like to ask whether there is a certain workflow that we have to follow to reconstruct structures like high antennas or generally type of models that have thin and long details. We're having big trouble in modelling them because of the noise that is being attached on the structures. There was an older thread about this matter which I cannot seem to be able to locate now. I'm attaching two screenshots of comparison between Metashape and another photogrammetric software. Keep in mind that we've tried masking the antenna, but there was no better result

[Alexey Pasumansky]

Hello cadm8,

Do you need to get the dense cloud or mesh of such structures? For mesh I can suggest to try depth maps based mesh generation.

If you have any example of the similar dataset that you can share with u s for internal testing, please send the download link to support@agisoft.com.

[Mak11]

Cadm8,

Here's the thread you were looking at I guess: https://www.agisoft.com/forum/index.php?topic=8984.15

As Alexey said if you don't need the dense cloud and you are only looking to build the mesh you should try the new depth maps based mesh generation workflow in Metashape.

Mak

[cadm8]

No Mak11 this is not the thread, it is the same dataset though. Alexey thank you, I need the point cloud for measuring, I'll provide you with the dataset used for the reconstruction.

[Alexey Pasumansky]

Hello cadm8,

Here's a first result of mesh reconstruction from the depth maps (in High quality), the result was not manually edited except for the cropping the surroundings and no masks were applied.

Is there any specific requirement, why you need the dense cloud and not mesh for the further work (for example, using application that deals with the point clouds only)?

[Alexey Pasumansky]

And here is the dense cloud generated from the same depth maps (High quality + Mild filtering) that were used for the mesh reconstruction. Also no editing except for the model cropping to filter out the surroundings.

[cadm8]

Alexey thank you for taking the time to check my dataset. I've rerun the data by realigning without GCPs, only exifs since that is what I sent you, in order to check it against your results, but I have to say that still the result from the point cloud isn't as clear as yours is. When your refer to "model cropping to filter out the surroundings", do you mean restricting the bounding box? As far as the point cloud, there are a lot of times that we need it for measuring reasons or overlaying data in general

[Alexey Pasumansky]

Hello cadm8,

I'll upload the project with the resulting mesh and dense cloud and share via PM, so that you can evaluate the output.