Geometry Based network simplification

Project Summary

Part of GSoC2022 / PySAL / momepy

The aim of this project was to explore different routes for simplifying a road network (mostly) for morphological analysis. My specific approcah was soley on geometry based operations. In summary, one PR has been merged into the main momepy code and has set the grounds of some further explorations; some of which have also made tier way into draft PRs that will be further developed past official end of GSOC2022. The accepted code was approached on the basis of developing a solution that is not only source agnostic, but also scallable. The main logic has been set by a combination of ideas by all parties; from the starting idea of polygonizing the network to the application of COINS algorithms, all the way to grouping adjacent error areas or even using the developed logic to rethinking remaining errors… it all has been a group effort by the whole team (see agknowledgements below).

Next steps are crucial for the completion of the end goal. Therefore, I have left some suggestions to mark a stepping stone of what it is expected to be a cause that would benefit quite a few, as demonstrated by the interest of many (see relevant discussion). Also, I provided a visual guide under the flowcharts section where I portrait a small landscape of the functions that could be developed

Agknowledgements

I would like express my special thanks of gratitude to Martin, James and Andres (mentors of the project), who from the beginning accommodated the project for an extra sit in what originally was a single person opportunity. I should also mention Gabriel (my peer contributor) with whom it was interesting collaborating coming from different backgrounds but sharing an objective. Their genuine patience to and constant virtual presence was vital for all my learnings. Additionally, they opened the doors to the whole PySAL dev community who I had been following for a long time in the darkness -to them also an earnestl acknowledgement.

Admittedly, I would have liked to have had more tangible results and contribute further with the project. Fortunately, as proven by many other PySAL developers, contributions during previous GSOC periods have only been the beginnings of something larger.

Repo file structure

This repository contains data, notebooks and general work in progress to the development of methods for road network simplifications.

/data : containing files used for testing.
/exploratory_notebooks : Jupyter Notebooks where the tests and debbugging where performed.
others : including licence, .py files (with self contained functions), cache etc.

Next Steps

As mentioned earlier, the main focus of this project was on a geometry based solution. Therefore, an obvious next step is to combine the results with solutions that are network conscious (the likes of those developed in parallel by @gagostini here or cityseer by Gareth Simons ).

However, there are still geometry solutions that would complement the methods merged up to this point. Currently momepy.roundabout_simplification() has two known issues:

Filtering out false negatives:
- Suggestion: consider using COINS as well as the current Circular Compactness to determine the roundabouts that are either not round enough or cut by other roads.
Not considering all adjacent polygons
- Suggestion: replace the current selection of adjacent roundabouts with one that uses the number of forming edges. This is likely to have a greater success rate given that most of those areas are ‘triangle-like’ and not necessarily smaller than the actual roundabout

Note: It is expected that this method need revising to leverage the vectorization advantages of Shapely 2.0

A natural transition for improvement could be to the full development of what could be summarised under the term “complex junctions”; i.e. junctions that in traffic the representation of street networks create additional nodes (and edges) that distor the results when doing morphological analysis. The suggested approach for dealing with some of these cases could be brielfy summarised with two main attributes:

Single & grouped polygons: resulting road network polygons that after classified as ‘invalid’ (see _selecting_invalid_polys() in PR #396 ) are either alone or touching other polygons
The number of forming edges: Different to exploding the polygons’ outer ring, this is an attribute of the number of edges that originally formed each invalid polygon. Said classification could help to identify different geometric solutions that would complement the toolbox for simplification.

Finally, one must mention that up to this point one of the most challenging problems to solve is the issue of parallel streets. For that, its suggested to investigate a route that combines the grouping of invalid polygons describes above with the voronoi centerline experiments by @martinfleis done previously. The hypothesis is that by creating a single polygon per group one could simplify their geometries into a single centerline that connectad the upcoming edges. All this is likely to be one of the last steps after solving the some of the other single issues.

Pull Requests (PR)

Flowcharts

Althought, quite certainly some processes and methods are likely to change, the following are some suggestions to achieve a simplified road network mainly using a geometry approach described above.

Detailed suggested flowchart

*_poly_center_line() : is expected to wrap/develop the processes teste by @martinfleis

Summary suggested flowchart

License

gregmaya/gsoc2022_network_simpl

ErrorLooks like something went wrong!

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Languages

Error
Looks like something went wrong!