Organized Point Filters

This module is a collection of filters for use on organized point clouds (OPC). Note that this software has not been as extensively tested as my other work. The filters:

• Laplacian Mesh Smoothing applied to an implicit fully connected right cut triangular mesh of an OPC.
  • Single threaded, CPU Multi-threaded, and GPU accelerated.
• Bilateral Mesh Normal Smoothing applied to an implicit fully connected right cut triangular mesh of an OPC.
  • Single threaded, CPU Multi-threaded, and GPU accelerated.
• Intel RealSense Bilateral Spatial and Disparity Transform filters used on depth images.
  • I thought it would be useful to pull this code out of the Intel SDK such that it can be used by others who are not using realsense cameras. Apache 2.0 License.

Here is an example GIF of Laplacian and Bilateral Filtering of a noisy organized pont cloud of stairs. The colors indicate the triangle normals of the mesh. The more uniform the colors, the smoother the surface

Installation

If you just want to install the python bindings and have a supported architecture, use pip: pip install polylidar. Binary wheels have been created for Windows, Linux, and MacOS. If this fails see below to install from source.

Build Python Library (Fast)

Building is entirely through CMake now. You must have CMake 3.14 or higher installed and a C++ compiler with C++ 14 or higher. CLone the project and then call pip install .. That its! It actually will call CMake for you and build the python library.

Build C++ Project Library

Building is entirely through CMake now. You must have CMake 3.14 or higher installed and a C++ compiler with C++ 14 or higher.

1. mkdir cmake-build && cd cmake-build. - create build folder directory
2. cmake ../ -DCMAKE_BUILD_TYPE=Release . For windows also add -DCMAKE_GENERATOR_PLATFORM=x64
3. cmake --build . -j$(nproc) --config Release - Build OPF

Build and Install Python Extension after C++ Library (Slow)

1. Install conda or create a python virtual envrionment (Why?). I recommend conda for Windows users.
2. cd cmake-build && cmake --build . --target python-package --config Release -j$(nproc)
3. cd lib/python_package && pip install -e .

If you want to run the examples then you need to install the following (from main directory):

1. pip install -r dev-requirements.txt

You also need cupy to be installed with cuda device drivers if you want GPU acceleration. I cant vouch that this will always work:

1. conda install cudatoolkit=10.1
2. pip install cupy-cuda101

Documentation

Please see documentation website for more details.

Citation

To support our work please cite:

@article{s20174819,
author = {Castagno, Jeremy and Atkins, Ella},
title = {Polylidar3D - Fast Polygon Extraction from 3D Data},
journal = {Sensors},
volume = {20},
year = {2020},
number = {17},
article-number = {4819},
url = {https://www.mdpi.com/1424-8220/20/17/4819},
issn = {1424-8220}
}

Bugs

A recent update to cupy is showing issues when using cuda for GPU filtering. When working on organized point clouds bigger than (430 X 430 X 3) the cupy error is: cupy_backends.cuda.api.driver.CUDADriverError: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered. This happened after the kernel is completed and trying to get the data from GPU to cpu. I have no idea why and don't really have the time to investigate. This error shows up on Windows 10 with RTX 2070 Super on March 1 2022.

Everything works superb for "smaller" organized point clouds however!