Proper LiDAR Classification Strategy

Question

Hello,

I am pretty new to working with LiDAR. Right now I have a LiDAR dataset that is unclassified beyond ground and bridges. I want to classify buildings and vegetation as well and I'm find the Classy Building and Classify by Height tools are pretty good. However, I'm running into scenarios where the edges of buildings are getting captured as vegetation, and I'm hoping there are some of you who could either suss out why that is, or point me to some useful tutorials (I've read through much of the ESRI-provided documentation for the tools).

Here's an example of what I mean:

The areas in the red circles are definitely parts of the buildings, but are being classified as medium vegetation. My workflow was first to run the Classify Building tool, and I mostly kept the defaults (2m minimum height and 4 sqm minimum area). That mostly does a good job, and is improved by enabling the classify above and below building options, but even that results in some edge errors (shown above), after running the Classify by Height and keeping the default settings of 5/25/50.

Thanks!

Ed_ · Answer

* Maybe first classify/segment vegetation using lidR package in R which is a very popular package in processing and analyzing LiDAR data. I know Pro has the Arc-R bridge, so you should be able to connect the two.
* https://github.com/redfoxgis/tree_segmentation this is a great example of classifying vegetation in an urban area using R and lidR. i was doing forestry LiDAR and vegetation analysis years ago and found this article really useful.

The lidR package

Working with Lidar Data in R

lidR classify by shapefile (merge_spatial) issues with user defined function

* If it works then move to classify buildings.
* Maybe look into Cloud Compare as well,  I haven't used in a while, but i know it's a good product. Maybe also look into Global Mapper, but it's not free.

ZachBodenner · Answer

Yes, I did actually already download the las tools for the decompression process, which was very useful, it went way faster than Pro. I didn't know there was a plugin, so this is helpful.

Okay let me prompt you one more time then: I have already run the classify buildings and classify by height tools, and what I wind up with is a lot of low vegetation (green dots):

One reason for this is that the classify by height tool doesn't have a method to exclude points below a certain height, so everything below my Low Vegetation threshold gets returned as such. That means minor (sub-foot) differences in points compared to ground (as seen mostly along the roadways there) and parked vehicles (many in the parking lot top-center).

From a best practices perspective, how would one go about addressing either of those issues without manual reclassification? I kind of feel like classifying by height to get vegetation categories should be the last thing to do since it's so broad and un-nuanced?

I'll play around with the las tools plugin, but figured I'd see if I maybe just did some steps out of order.

Ed_ · Answer

Yeah manually removing noise can be time consuming just like digitization lol

idk if you are familiar with LAS tools but it's very famous in the world of processing LiDAR data via open source. It does seem to have a plugin for Pro.

LAStools ArcGIS Pro Plugin Manual - rapidlasso GmbH

Maybe try the Las Height tool using the Iterate Files tool in Model Builder

Global Mapper Pro is another famous software product outside ESRI that's good with preprocessing LiDAR data.

Sometimes LAS files also have noise classified as 7 and/or 8

If you are comfortable with R then it also has some really good LiDAR related libraries/packages.