Segmentation of UAV point cloud

The point cloud contains a row of continuous trees. I want to segment individual trees. Kindly help. I tried it by distance, there are death trees inbetween.

[Benjamin Thompson]

Can you post the raw data, and more information about how to distinguish one tree from another? Basically what should the output of this "segmentation" look like for a small example?

[Mugilan Govindasamy Raman]

I tried uploading the raw file, I am unable to upload considering the size.

The point cloud contains a row of (approx. 100) trees which includes dead trees and poles in between. I am trying to segment individual trees to extract the features such as height, volume and branching information for each individual trees.

[Benjamin Thompson]

Upload a smaller portion of the data set. Otherwise the community can not look into the problem in any more detail then the images you have already loaded.

[Mugilan Govindasamy Raman]

Kindly see the attachment. I have attached both .las and .csv format.

[Benjamin Thompson]

Can you define the data in the IAS and CSV files? What does each column mean?

[Mugilan Govindasamy Raman]

The las and csv has three column representing X, Y, Z. The X and Y represents the positions (Latitude and Longitude in meters) and Z represents the elevation (Height). The data is in projected coordiante system. WGS 1984 UTM Zone 10N.

Answer 1

[Githin George]

Hello,

I understand you have point cloud data for a row of trees and would like to segment the individual trees and extract their attributes.

I’ve attached an example which describes the segmentation process below.

Extract Forest Metrics and Individual Tree Attributes from Aerial Lidar Data - MATLAB & Simulink (mathworks.com)

Note that you will have to convert the unorganized point cloud data to organized point cloud.

In case you would like to use the unorganized pcd, you can try the following steps as a workaround.

1. Use statistical measures like ‘min’ of data to manually remove ground data points and apply “pcsegdist” function to get better segmentation results.
2. Classify the segments into dead trees by comparing statistics like ‘mean’ of height, or ‘std variance’ of width.
3. In case using “pcsegdist” once is not yielding good results, you could try applying it again on the obtained segments but with finer value of ‘minDistance’ parameter.

Please refer to the following documentation link for further information on “pcsegdist” function.

https://mathworks.com/help/vision/ref/pcsegdist.html#d126e269205

I hope this helps.