Roof planes extending much farther and higher than what can be reasonably assumed from the data

[isak-rh]

Sometimes roofer will extend a roof plane into areas with no data in a way that makes the final building much higher than the data would reasonably suggest. I see this happening a fair amount with the Lantmäteriet dataset I'm working with (I'm the person who posted #169, just posting from my corporate account now.) The most common case is small single family homes with gable roofs sometimes only end up with coverage on one side of the roof, and roofer will extend the plane to cover the entire roof, making the final building much taller than it should be.

The 3DBAG seems to be affected by this as well. It didn't take me long to find the following in Ruurlo (I figured I should look in a small town.) Google maps confirms that these should be normal gable roofs:

#170 probably goes some way towards solving this but I think there might be more fine grained ways to do a sanity check. I've seen buildings with less than 50% coverage come out fine provided that coverage is distributed well enough. Checking that a building does not extend significantly higher than measured might be a good way of catching this and is something I've been experimenting with this in my post-processing workflows.

[Ylannl]

This a known problem, nevertheless thank you for opening an issue on this, we should try to address it and this issue will help motivate us.

The issue can be caused by not enough points in certain areas, but also by a failure to detect a line segment that is needed to properly split to parts of the roof into separate faces. The latter is probably what is happening in the cases you describe.

[isak-rh]

Thanks for checking it out, good to know that it's already a known issue. I can't be sure what's going on in the 3DBAG example, but in the Lantmäteriet data I'm working with I mostly see the case I'm mentioning, see this example:

But I do also see cases that might be consistent with what you describe, where there are enough points on the more obscured side that a plane should be detected but with more sparse and irregular coverage.

For what we're doing I will just handle the fallback in post but I figured it was worth opening an issue.

[Ylannl]

Ah, indeed you are right. Thanks for the added context.