We have been implementing python binding to the pointcloud <http://pointclouds.org/>_ library (PCL). This is a work in progress and currently the following modules/functionalities of the API are wrapped:
- I/O and integration; saving and loading PCD (Point Cloud Data) files.
- segmentation
- SAC (Simple Alignment Consensus)
- smoothing
- filtering
- registration
We aim to wrap the following modules/functionalities:
- feature extraction
- ....
Note that all methods operate in Cartesian coordinate system i.e. (x, y, z). Point cloud data comes in two formats (x,y,z) and (x,y,z, R, G, B). These methods work on both.
some useful links to PCL:
This release has been tested on:
Ubuntu 13.10 with,
- Python 2.7.5
- pcl 1.7.1
- Cython 0.20.2
CentOS 6.5 with,
- Python 2.6.6
- pcl 1.6.0
- Cython 0.20.2
Fedora 20 with,
- Python 2.7.5
- pcl 1.7.0
- Cython 0.21.1
The instruction below holds for Ubuntu and Fedora.
You need to have the following packages installed on your system:
In Ubuntu
sudo apt-get install cmake libgdal-dev pcl
sudo apt-get install cmake libgdal-dev libgeotiff-dev
In Fedora
System-wide:
sudo yum install gdal-devel libgeotiff-develsudo yum install boost-develsudo yum install gcc-c++
Python libraries:
sudo yum install python-pip python-develsudo pip install numpysudo pip install scipy
LibLAS:
git clone git://github.com/libLAS/libLAS.git liblas
cd liblas/
mkdir makefiles
cd makefiles/
cmake -G "Unix Makefiles" ../ --GEOTIFF_INCLUDE_DIR=/usr/include/libgeotiff/
make
sudo make install
NOTE: You might need to export LD_LIBRARY_PATH i.e in your .bashrc file, add this line:
export LD_LIBRARY_PATH=/usr/local/lib
PCL:
sudo yum install pcl pcl-devel pcl-tools pcl-doc
Clone repository, compile and test that it works:
git clone 'http://github.com/larsmans/python-pcl'
cd python-pcl/
make
python -m unittest discover testsNow you should be able to use python-pcl modules and functionalities as explained below.
It instantiates a point could object. It also provides functionality for manipulating point cloud data in a numpy array.
pcl.PointCloud()
pcl.PointCloudXYZRGB()
example
import numpy as np
import pcl
arr = np.zeros((100, 3)).astype(np.float32)
pc = pcl.PointCloud()
pc.from_array(arr)
pc.to_array()
import numpy as np
import pcl
arr = np.zeros((100, 6)).astype(np.float32)
pc = pcl.PointCloudXYZRGB()
pc.from_array(arr)
pc.to_array()
It helps loading and saving point cloud files.
pcl.load(path, format=None, loadRGB=False)
pcl.save(cloud, path, format=None, binary=False)
example
Load a point cloud from PCD file:
import pcl
pc = pcl.load('tests/bun0.pcd')
Load a point cloud from PLY file:
import pcl
pc = pcl.load('tests/rock2.ply')
Load a point cloud from PLY file
import pcl
pc = pcl.load('tests/rock2.ply', loadRGB=True)
Save point cloud to PLY file (save to PCD is also supported).
pcl.save(pc, 'tests/tmp.ply', format='PLY')
pcl.Segmentation()
...
example
import pcl
pc = pcl.load('some_cloud.pcd')
seg = pc.make_segmenter()
seg.set_optimize_coefficients (True)
seg.set_model_type(pcl.SACMODEL_PLANE)
seg.set_method_type(pcl.SAC_RANSAC)
seg.set_distance_threshold (0.01)
indices, model = seg.segment()
pc2 = pcl.PointCloud()
pc2.from_array(pc.to_array()[indices,:])
pcl.save(pc2, 'other_cloud.ply', format='PLY')
import pcl
pc = pcl.load("some_cloud.pcd")
seg = pc.make_segmenter_normals(50)
seg.set_optimize_coefficients (True);
seg.set_model_type (pcl.SACMODEL_CYLINDER)
seg.set_method_type (pcl.SAC_RANSAC)
seg.set_normal_distance_weight (0.1)
seg.set_max_iterations (10000)
seg.set_distance_threshold (0.05)
seg.set_radius_limits (0, 0.1)
indices, model = seg.segment()
pc2 = pcl.PointCloud() # to get a second point cloud of the segmented point cloud
pc2.from_array(pc.to_array()[indices,:])
pcl.save(pc2, 'other_cloud.ply', format='PLY')
example
smoothing does ... and it is implemented as ....
example
import pcl
p = pcl.load("C/table_scene_lms400.pcd")
fil = p.make_statistical_outlier_filter()
fil.set_mean_k (50)
fil.set_std_dev_mul_thresh (1.0)
fil.filter().to_file("inliers.pcd")
...
example
In registration module, we have wrapped four registration method ICP (Iterative Closest Point), generalized ICP, generalized non-linear ICP, initial alignment RANSAC ():
pcl.registration.icp(BasePointCloud source, BasePointCloud target, max_iter = None)
pcl.registartion.gicp(BasePointCloud source, BasePointCloud target, max_iter = None)
pcl.registartion.icp_nl(BasePointCloud source, BasePointCloud target, max_iter = None)
pcl.registartion.ia_ransac(BasePointCloud source, BasePointCloud target, max_iter = None, radius=0.05, minSampleDistance=0.05, maxCorrespondenceDistance=0.2)
example
import pcl
import pcl.registration
sourceFile = 'tests/color.ply'
targetFile = 'tests/color.ply'
source = pcl.load(sourceFile)
target = pcl.load(targetFile)
converged, transf, estimate, fitness = pcl.registration.icp(source, target)
import pcl
import pcl.registration
sourceFile = 'tests/color.ply'
targetFile = 'tests/color.ply'
source = pcl.load(sourceFile)
target = pcl.load(targetFile)
converged, transf, estimate, fitness = pcl.registration.gicp(source, target)
import pcl
import pcl.registration
sourceFile = 'tests/color.ply'
targetFile = 'tests/color.ply'
source = pcl.load(sourceFile)
target = pcl.load(targetFile)
converged, transf, estimate, fitness = pcl.registration.icp_nl(source, target)
import pcl
import pcl.registration
sourceFile = 'tests/color.ply'
targetFile = 'tests/color.ply'
source = pcl.load(sourceFile)
target = pcl.load(targetFile)
converged, transf, estimate, fitness = pcl.registration.ia_ransac(source, target)
numpy helper does ...
example
import pcl
import numpy as np
p = pcl.PointCloud(np.array([[1, 2, 3], [3, 4, 5]], dtype=np.float32))
seg = p.make_segmenter()
seg.set_model_type(pcl.SACMODEL_PLANE)
seg.set_method_type(pcl.SAC_RANSAC)
indices, model = seg.segment()
.. autosummary:: pcl.PointCloud pcl.Segmentation pcl.SegmentationNormal pcl.StatisticalOutlierRemovalFilter pcl.MovingLeastSquares pcl.PassThroughFilter pcl.VoxelGridFilter