This Python repository hosts the final source code related to my MSc thesis at the TU Delft, titled "Constructing a digital 3D road network for The Netherlands". The research proposal and the thesis itself are based in my LaTeX repository.
All planned features of the software have been implemented and tested, and the project has been concluded. The final thesis is now available in the LaTeX repository. Only minor modifications or additions are expected to the code from now on.
I packaged most functionality into one "module" called nbrs_generation.py. Almost all the features reside in the class nbrs_manager, to provide an interface that is easy to get familiar with for potential re-users. A set of processing functions were factored out into lib_shared.py to reduce clutter in the main module.
The documentation of the software is "inline", i.e. a comprehensive set of docstrings and comments are provided in the code. For a top-level description of how the interface is meant to be used, please see the docstring of the nbrs_manager class.
Sample inputs and outputs can be downloaded from here. They were all generated using the example code provided at the end of nbrs_generation.py (the code that runs if the file itself is run).
The files were compressed into a 7-Zip file (not conventional Zip), because of their considerable size. The size of the 7-Zip archive is about 2Gb. All inputs and outputs (including intermediate results) are included for all testing datasets listed in Section 4.1.2 of the P4 report. Output files based on active contour optimisation are only included for the datasets C_20BN1, C_37HN2 and C_39CZ1. Outputs tagged _conts at the end of their file names are active contour optimisation-based versions of their normal counterparts. To visualise attractor maps, I recommend using minimum and maximum bounds of 0.998 and 1 respectively.
Note that some LineStrings in the output may have all-NaN elevations and empty Lidar subclouds. This is due to issues with how the Lidar input files were produced; some NBRS extend beyond the boundaries of the relevant AHN3 tiles and can therefore not be associated with elevations. This issue bears no relevance to the implementation's effectiveness.
This work is licensed under an MIT license.