Python bindings for the Point Cloud Library
Project description
pclpy: PCL for python
Python bindings for the Point Cloud Library (PCL). Generated from headers using CppHeaderParser and pybind11.
This library is in active development, the api is likely to change. The included modules do work, but tests are incomplete, and corner cases are still common.
Only Windows and python 3.6 x64 are supported at the moment.
Contributions, issues, comments are welcome!
Github repository: https://www.github.com/davidcaron/pclpy
Pypi: https://pypi.org/project/pclpy/
Motivation
Many other python libraries tried to bind PCL. The most popular one being python-pcl, which uses Cython. While Cython is really powerful, binding C++ templates isn't one of its strenghts (and PCL uses templates heavily). The result for python-pcl is a lot of code repetition, which is hard to maintain and to add features to, and incomplete bindings of PCL's classes and point types.
Using pybind11, we use C++ directly. Templates, boost::smart_ptr and the buffer protocol are examples of things that are simpler to implement.
The results so far are very promising. A large percentage of PCL is covered.
Installing
Windows with python 3.6 x64
pip install pclpy
When pip installs the project, pclpy_dependencies
is installed as a requirement.
This simple package contains only the PCL dlls required on Windows so you don't have
to download a PCL release or build it.
Linux
Not working for now. Contributions are welcome!
Features
- All point types are implemented (those specified by the default msvc compile flags)
- You can view point cloud data as numpy arrays using
cloud.x
orcloud.xyz
- boost::shared_ptr is handled by pybind11 so it's completely abstracted at the python level
- laspy integration for reading/writing las files
Example
You can use either a high level, more pythonic api, or the wrapper over the PCL api. The wrapper is meant to be as close as possible to the original PCL C++ api.
Here is how you would use the library to process Moving Least Squares. See the PCL documentation: http://pointclouds.org/documentation/tutorials/resampling.php
Using the higher level api:
import pclpy
# read a las file
point_cloud = pclpy.read("street.las", "PointXYZRGBA")
# compute mls
output = point_cloud.moving_least_squares(search_radius=0.05, compute_normals=True, num_threads=8)
Or the wrapper over the PCL api:
import pclpy
from pclpy import pcl
point_cloud = pclpy.read("street.las", "PointXYZRGBA")
mls = pcl.surface.MovingLeastSquaresOMP.PointXYZRGBA_PointNormal()
tree = pcl.search.KdTree.PointXYZRGBA()
mls.setSearchRadius(0.05)
mls.setPolynomialFit(False)
mls.setNumberOfThreads(12)
mls.setInputCloud(point_cloud)
mls.setSearchMethod(tree)
mls.setComputeNormals(True)
output = pcl.PointCloud.PointNormal()
mls.process(output)
You can see the wrapper is very close to the C++ version:
// C++ version
pcl::PointCloud<pcl::PointXYZ>::Ptr point_cloud (new pcl::PointCloud<pcl::PointXYZ> ());
pcl::io::loadPCDFile ("bunny.pcd", *point_cloud);
pcl::MovingLeastSquaresOMP<pcl::PointXYZ, pcl::PointNormal> mls;
pcl::search::KdTree<pcl::PointXYZ>::Ptr tree (new pcl::search::KdTree<pcl::PointXYZ>);
mls.setSearchRadius (0.05);
mls.setPolynomialFit (false);
mls.setNumberOfThreads (12);
mls.setInputCloud (point_cloud);
mls.setSearchMethod (tree);
mls.setComputeNormals (true);
pcl::PointCloud<pcl::PointNormal> output;
mls.process (output);
Modules
- 2d
- common
- geometry
- features
- filters
- io
- kdtree
- keypoints
- octree
- recognition
- sample_consensus
- search
- segmentation
- stereo
- surface
- tracking
- visualization
These modules are skipped for now
- ml
- people
- outofcore
- registration
- every module not in the PCL Windows release (gpu, cuda, etc.)
Not Implemented
(see github issues
and the what to skip section in generators/config.py
)
To build
Windows with python 3.6 x64
- Download PCL release for Windows (PCL-1.8.1-AllInOne-msvc2017-win64.exe) at: https://github.com/PointCloudLibrary/pcl/releases/download/pcl-1.8.1/PCL-1.8.1-AllInOne-msvc2017-win64.exe
- PCL_ROOT environment variable must be set to the installation directory of PCL
- About requirements:
- Install pybind11 from github (2.3dev version) it includes a necessary bug fix
- Install CppHeaderParser from https://github.com/davidcaron/CppHeaderParser (specific bug fixes)
- Generate modules using
generate_pybind11_bindings.py
- There is a missing file from the PCL release that you should get from the github repo: 2d/impl/kernel.hpp
- Must be built with x64 version of cl.exe because of the large memory usage (see workaround in setup.py)
- python setup.py install
- Useful setup.py arguments:
- --msvc-mp-build should enable a multiprocessed build
- --msvc-no-code-link makes linking much faster (do not use for releases, see setup.py description)
- --use-clcache to cache msvc builds using clcache (must be installed)
- --debug to build in debug mode
Roadmap
- Wrap as much of PCL as reasonably possible
- More tests
- CI on Appveyor
- Make it work on Linux
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distribution
Hashes for pclpy-0.11.0-cp36-cp36m-win_amd64.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 0dbc082e86723a267fe423bda65866783b4664a8e0b1b8d6305df8c839a38b85 |
|
MD5 | 18324f46c9a5bec532dc81408e6e282c |
|
BLAKE2b-256 | e1097f174c4256e56122eec468782aa3a5cc3bd9f8f6540635efa19d10d7897c |