Extract geometry features from 3D point clouds

Content

The project contains the following folders:

• geo3dfeatures contains source code
• docs contains some mardown files for documentation purpose and images
• examples contains some Jupyter notebooks for describing data
• tests; pytest is used to launch several tests from this folder

Additionally, running the code may generate extra subdirectories in a chosen data repository (./data, by default).

How to install

This projects runs with Python3, every dependencies are managed through poetry.

Installation from source

$ git clone ssh://git@git.oslandia.net:10022/Oslandia-data/geo3dfeatures.git
$ cd geo3dfeatures
$ virtualenv -p /usr/bin/python3 venv
$ source venv/bin/activate
(venv)$ poetry install

Contribution

See CONTRIBUTING.md.

Run commands

In order to get the available program commands, consider the program help (geo3d -h):

usage: geo3d [-h] {info,sample,index,featurize,cluster,train,predict} ...

Geo3dfeatures framework for 3D semantic analysis

positional arguments:
  {info,sample,index,featurize,cluster,train,predict}
    info                Describe an input .las file
    sample              Extract a sample of a .las file
    index               Index a point cloud file and serialize it
    featurize           Extract the geometric feature associated to 3D points
    cluster             Cluster a set of 3D points with a k-means algorithm
    train               Train a semantic segmentation model
    predict             Predict 3D point semantic class starting from a
                        trained model

optional arguments:
  -h, --help            show this help message and exit

Any further CLI documentation may be printed with geo3d <command> -h.

Documentation

Some documentation is available, that describes the set of considered geometric features, the fixtures (i.e. dummy datasets) used for test purpose and a practical pipeline use case:

• Feature set
• Test fixtures
• Command pipeline

Examples

The following example has been generated starting from a CANUPO dataset (file scene.xyz, with 500k points, 50 neighbors and all the features):

Oslandia – 2019-2020