Issues Classifying Ground Points

[FrostBite]

I'm trying to classify ground points to make a high quality DTM. For some reason the 'Classify Ground Points' function doesn't get all points on the ground. Some rocks and logs that have a steep angle don't register as ground. The next thing I would do is select all the unclassified points and call them high vegetation (since this is a forested area), but this would mean that some of the ground would be also called high vegetation.

I've tried a few different iterations but I still haven't achieved a good result.

What do you all do in this situation? Does it really matter that much? I am not sure what the best course of action is!

I attached a screenshot of part of the ground.

[vineg]

Hi!

There is a tweak that allows to add points near ground surface to the ground class.

You can set it in Tools->Preferences menu, Advanced tab, Tweaks... button

Parameter name is ClassifyGroundPoints/buffer_width and value is height above ground in meters.
Logs and rocks smaller than this value will be classified as ground. So you could set it to 0.4 for example.

As a side effect lower parts of high vegetation will be added to the ground as well, setting "Erosion radius" to the same value as buffer_width can help to avoid it.

Regards, Egor

[FrostBite]

Thanks for the reply Egor!

This did help quite a bit! But now the smaller trees, 10-15m tall, are also classified as ground.

It would help with understanding what your settings are doing. And I haven't quite wrapped my head around cell size and max angle...

[vineg]

Hi!

Here is an article https://agisoft.freshdesk.com/support/solutions/articles/31000160729-parameters-for-ground-point-classification
and an algorithm reference https://www.isprs.org/proceedings/Xxxiii/congress/part4/111_XXXIII-part4.pdf

Basically we initially split point cloud to cells with the specified size and looking for initial ground points in each cell. So each cell should be big enough to contain at least 1 ground point.

Max angle controls how smooth final ground points will be. Low value - almost flat, high - more detailed/noisy. Usually 15 works good.

About tweak, I rechecked, it appears to use local scaled coordinates instead of meters. Try to lower it to a factor of 10.
We will fix it to use meters in the next 2.2.0 release.

[FrostBite]

I'm still not able to properly classify ground points.
For the most part it worked better, but ground points that may be stumps, rocks or a small mound, for example, will only be half classified as ground.

Another issue is that in the forest there are fallen trees that are included as ground (which is ok), but then a large portion will be unclassified.

The problem with that, is I normally classified ground points, and then all the other unclassified points (which should be trees, since I'm mapping forests) would be high vegetation. But if there are unclassified points on the ground and I call them high vegetation, it would create a lower quality Ortho. Maybe this is a seperate problem I'm having...

I attached a picture of the Ortho.

I have no idea what I'm doing wrong with my processing. In previous versions I had a lot less issues. But perhaps it's because I have a new drone (Mavic 3 Enterprise) and a new computer. Is it because I'm processing the dense cloud at Ultra High settings?

[Alexey Pasumansky]

Hello FrostBite,

If you have problematic project, you can share it (with alignment results and classified point cloud) to support@agisoft.com and also specify the classification parameters that you are using, so that we could see, if the classification result can be improved by using different parameters or whether there is any problem in the classification algorithm.