Discussion: Classifying lidar

Hi AP community.

Does anyone have any experience classifying lidar data and might be able to help me out? I have a set of lidar laz files that are only classified as ground and non-ground and neither KP nor OCAD were able to produce anything using their vegetation algorithms. I'm assuming the best way forward is to further classify the non-ground data but I've never done that before. Any pointers would be appreciated.

Are you sure you have non-ground points there and not ground points only? I am asking because KP sees all points other than ground or water as vegetation, so classifying non-ground points as vegetation will make no difference. try drag-dropping the laz file on top of lasview and see how it looks, mouse right click and color by classification, also try render only ground / vegetation etc may tell you whats going on.

I'll try that again. I definitely got really messed up output but I'm actually not 100% sure the veg didn't work. I do remember getting cliffs data that was very clearly wrong. I think every individual tree was shown as being full of cliffs. This was back in January/February when I first processed the data.

I had once a file, where for a number of the tiles, not all, the published point cloud had for some reason no ground points (Jordan figured that out for me). I was lucky that they had also published a "grounds only" file, so using LAStools I was able to merge the files, eliminate duplicates and then I had data that worked properly. Not saying that is the case, but who knows.

Another thing to check is the scale/coordinate system, if that is wrong (basically not in meters) then the tools will have hard time figuring out vegetation.

Sounds like it isn't classified. Reprocess using lasground_new.
https://github.com/LAStools/LAStools/blob/master/b...

Yeah, you start from scratch, assuming no or faulty classification, then run lasground_new followed by steps to classify buildings, then split the remaining (vegetation) points into low/medium/high.
Depending on lidar density and quality you might have to experiment with the various lasground options, like town/city/etc.

I've also used lasground_new for general classification, but I tend to use lasground because of this video, where Martin says lasground_new was written for a specific case of an urban area next to mountains. The case that made him create lasground_new was an area in the Philippines.

It's near the beginning of this video:

https://www.youtube.com/watch?v=NlwkSv0BuQU&li...

If you read the lasground_new readme file, it only says, "This is a totally redesigned version of lasground that handles complicated terrain much better where there are steep mountains nearby urban areas with many buildings."

But it doesn't say to not use it elsewhere. So it's confusing, and I'm not 100% sure myself.

I had two really difficult "large tree" areas where I spent significant time reclassifying the lidar. Big Basin Redwoods State Park in California had remarkably few ground points under the dense canopy areas. I don't think reclassiying helped much. The other area I had big trouble with was a large tree area north of Wellington, NZ. The problem there was that the limits of the free version of lastools prevented me from improving the original classification. I had to tile at 250m or 500m, and lasground really needed to process across the whole area to produce good results. I can't remember if I accepted the reduction in data lastools imposes when you exceed the point count. I couldn't improve on the original classification done by the professionals. Usually if you have a few buildings classified as ground you can improve on it.

The other place where the free limits were impossible to work around was classifyng ground points in the "dense point cloud" output of drone mapping outputs. There are hundreds or thousands of points per square meter, and you're stuck with a ~250m tile and a bounding box limit that prevents you from having good results---but only in the areas with forest. In lawn areas with trees that don't touch, the results were very good. (I don't remember exactly, but I recall that I used a 300m tile with a 100m bounding box, so only a 100m x 100m tile was the output of running each 300m x 300m tile, and the run time was very long because 8/9ths of each tile was clipped. I think you specify the "clipped size" of the tile and an additional bounding box "border", and lastools won't let you specify a bounding box bigger than the inside. I wanted to run a 64-m tile with a 128m bounding box, and it won't let you. I think I was using a 25m or 50m step, and my theory was you wanted at least 3x the step size as the bounding box---so if you used a step of 50m, you'd want a bounding box of at least 150m. I think I just accepted the loss of data and processed it as one big file.

@cedarcreek: I didn't notice the Philippines connection but I have in fact used lasground_new in that country, with no real issues.

My most complicated project was when I created my own maps of a couple of areas near Santa Cruz, California: Default parameters did in fact lead to contours around each of the big trees, but just increasing the step size a few times gave me a totally useable set of contours: Not nearly as good as what I get from 5-10 pulses/square meter here in Norway, but the actual contour detail is far less complicated anyway.
https://tmsw.no/qr/show_map.php?user=terjem&ma...
https://tmsw.no/qr/show_map.php?user=terjem&ma...