Using MCC-RGB and MCC

This is a simple example of applying pymccrgb’s ground classification algorithm to a multispectral lidar dataset with colors at each point. It is compared to the MCC algorithm devloped by Evans and Hudak, 2007.

[1]:

import warnings
warnings.filterwarnings('ignore')

[2]:

from pymccrgb.core import mcc, mcc_rgb
from pymccrgb.plotting import plot_points_3d, plot_results
from pymccrgb.datasets import load_mammoth_lidar

First, we load the dataset, a lidar point cloud over Horseshoe Lake near Mammoth Mountain, CA. This was acquired by the National Center for Airborne Laser Mapping with an Optech Titan three-channel lidar scanner.

[3]:

data = load_mammoth_lidar(npoints=1e6)

[9]:

plot_points_3d(data)
../_images/examples_mcc-rgb_6_0.png

We can use the standard MCC algorithm (mcc) to classify the points by relative height.

[5]:

data_mcc, labels_mcc = mcc(data, verbose=True)

--------------------
MCC iteration
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 0
Removed 314976 nonground points (31.50 %)
--------------------
MCC iteration
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 1
Removed 90793 nonground points (13.25 %)
--------------------
MCC iteration
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 2
Removed 24723 nonground points (4.16 %)
--------------------
MCC iteration
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 3
Removed 7501 nonground points (1.32 %)
--------------------
MCC iteration
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 4
Removed 3361 nonground points (0.60 %)
--------------------
MCC iteration
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 0
Removed 10577 nonground points (1.89 %)
--------------------
MCC iteration
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 1
Removed 2768 nonground points (0.51 %)
--------------------
MCC iteration
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 2
Removed 1502 nonground points (0.28 %)
--------------------
MCC iteration
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 3
Removed 177 nonground points (0.03 %)
--------------------
MCC iteration
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 0
Removed 12893 nonground points (2.37 %)
--------------------
MCC iteration
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 1
Removed 3545 nonground points (0.67 %)
--------------------
MCC iteration
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 2
Removed 1678 nonground points (0.32 %)
--------------------
MCC iteration
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 3
Removed 130 nonground points (0.02 %)
--------------------
MCC iteration
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 4
Removed 16 nonground points (0.00 %)
Retained 525360 ground points (52.54 %)

… or use MCC-RGB (mcc_rgb) to classify points by relative height and color.

To do this, we specify a training height tolerance. In this case, points over 0.3 m relative height will be considered vegetation (non-ground) points. These points are used to train a color-based classifer that is used to re-classify low non ground points.

[6]:

data_new, labels_new = mcc_rgb(data, verbose=True)

--------------------
MCC step
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 0
Removed 314976 nonground points (31.50 %)
--------------------
Classification update step
--------------------
Scale: 0.50, Relative height: 3.0e-01
Removed 168135 nonground points (16.81 %)
--------------------
MCC step
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 1
Removed 10830 nonground points (1.08 %)
--------------------
MCC step
--------------------
Scale: 0.50, Relative height: 3.0e-01, iter: 2
Removed 2340 nonground points (0.23 %)
--------------------
MCC step
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 0
Removed 8688 nonground points (0.87 %)
--------------------
MCC step
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 1
Removed 2465 nonground points (0.25 %)
--------------------
MCC step
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 2
Removed 1321 nonground points (0.13 %)
--------------------
MCC step
--------------------
Scale: 1.00, Relative height: 3.0e-01, iter: 3
Removed 110 nonground points (0.01 %)
--------------------
MCC step
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 0
Removed 11724 nonground points (1.17 %)
--------------------
MCC step
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 1
Removed 3258 nonground points (0.33 %)
--------------------
MCC step
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 2
Removed 1489 nonground points (0.15 %)
--------------------
MCC step
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 3
Removed 132 nonground points (0.01 %)
--------------------
MCC step
--------------------
Scale: 1.50, Relative height: 3.0e-01, iter: 4
Removed 10 nonground points (0.00 %)
Retained 474522 ground points (47.45 %)

Finally the plot_results convenience function compares the final results of each algorithm visually.

[7]:

plot_results(data, labels_mcc, labels_new)
../_images/examples_mcc-rgb_12_0.png