Methods for identifying individual trees from point cloud data
Segments a point cloud from a LAS file into individual trees.
This script works with Python 3. It takes as input a LAS file and outputs two csvs assigning the points from the LAS file into clusters, one cluster per tree plus one cluster for the ground. In our testing it clustered ~6000 points in ~16 seconds, with runtimes increasing as O(n^2) in the number of points.
There are 3 steps to the script:
- Identifying "stems", which are a list of points with descending z-value, that go from the top of a tree to the ground. In the final clustering, each stem will generate one cluster, which should consist of one tree.
- Do an initial clustering pass by "growing" each stem. For each point in the cluster, we add the unclustered points in a cylinder above it to the same cluster, repeating until this adds no more new points to the cluster.
- Cluster the remaining points by finding its nearest neighbor that was clustered in step (2) and assigning it to the same cluster.
For each input file, there are two output files (you can also plot the outputs, see below):
- In cluster_csvs/ the files ending in _clusters will be a csv file with one row per point, and columns index, x,y, z, height_above_ground, and cluster. Cluster will be an integer, with 0 denoting the ground, each positive number denoting a separate tree, and -1 denoting any unclustered points (if there are any)
- In cluster_csvs/ the files ending in _stem_centers will be a csv file with one row per cluster (usually 1 row for the ground and one row per tree), and columns index, x, y, z, height_above_ground, cluster, and cluster_name. The x,y,z coordinates represent the center of the cluster.
Please don't hesitate to open an Issue if you find any problem or suggestions for a new feature.
Before using this script, you must install laspy:
pip3 install laspy
You can run the script from the command line as python3 tree_segment.py. You will likely want to use optional arguments to set the input file and some configuration options. You can see the optional arguments by running python3 tree_segment.py -h:
usage: tree_segment.py [-h] [-f FILE_NAME] [-td] [-sv SIDE_VIEW] [-dh DESCENDENTS_HEIGHT] [-of OUTPUT_FOLDER]
Clusters points from a .las file into trees.
optional arguments:
-h, --help show this help message and exit
-f FILE_NAME, --file_name FILE_NAME
the name of the input file
-td, --top_down whether to save a top-down plot of the clustering results
-sv SIDE_VIEW, --side_view SIDE_VIEW
whether to save a side-view plot of the clustering results, and at what angle to view it. must be >0 (default: 0)
-dh DESCENDENTS_HEIGHT, --descendents_height DESCENDENTS_HEIGHT
how far to search vertically for descendents when searching for stems (default: .3m)
-dr DESCENDENTS_RADIUS, --descendents_radius DESCENDENTS_RADIUS
how far to search horizontally for descendents when searching for stems (default: .2m)
-msh MINIMUM_STEM_HEIGHT, --minimum_stem_height MINIMUM_STEM_HEIGHT
minimum height required for a stem to be valid (default: .5m)
-fsdr FOUND_STEM_DISABLE_RADIUS, --found_stem_disable_radius FOUND_STEM_DISABLE_RADIUS
points within this distance of a found stem will be disabled (default: .7m)
-ftfdr FAIL_TO_FIND_DISABLE_RADIUS, --fail_to_find_disable_radius FAIL_TO_FIND_DISABLE_RADIUS
points within this distance of a point with no path to the ground will be disabled (default: .2m)
-gr GROW_RADIUS, --grow_radius GROW_RADIUS
how far to search horizontally for other points in the tree when "growing" (default: .2m)
-gh GROW_HEIGHT, --grow_height GROW_HEIGHT
how far to search vertically for other points in the tree when "growing" (default: .4)
-ghc GROUND_HEIGHT_CUTOFF, --ground_height_cutoff GROUND_HEIGHT_CUTOFF
points below this height will be clustered into the ground cluster (default: .2m)
-of OUTPUT_FOLDER, --output_folder OUTPUT_FOLDER
the name of the folder where csv outputs are saved
Example 1: Use default parameters to cluster the points in sample_data/treeID_40645_merged.las:
python3 tree_segment.py
# which is equivalent to
python3 tree_segment.py -f sample_data/treeID_40645_merged.las
# or
python3 tree_segment.py --file_name sample_data/treeID_40645_merged.las
Example 2: Saving plots with a top-down perspective:
python3 tree_segment.py -f sample_data/treeID_40645_merged.las -td
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -td
# these should appear in the saved_images/ folder
Example 3: Saving plots from a side-view perspective:
# create a side view of the trees rotated 90 degrees:
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -sv 45
# these should appear in the saved_images/side_views/ folder
# the image is somewhat unclear because a the angle makes two trees overlap. let's try this angle instead:
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -sv 135
# that's clearer! some other angles:
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -sv 180
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -sv 185
python3 tree_segment.py -f sample_data/treeID_40645_merged.las -sv 360
Example 4: Saving the output csvs to another folder:
# saves the csv outputs to the folder example_folder/
python3 tree_segment.py -f sample_data/treeID_40113_merged.las -of example_folder
- Code: Robert Huben
- Data and testing: Zhengyang Wang