OTP dataloading runs an integrated gtfs-osm stop matching task defined in https://github.com/HSLdevcom/OpenTripPlanner-data-container

Fits GTFS shape files and stops to a given OSM map file. Uses pymapmatch for the matching. Usually quite accurate.

Copyright 2013 and 2014 Jami Pekkanen (with dots at helsinki.fi), released under AGPLv3 license.

• Linux. May work on other unix-like systems. Don't know about Windows.
• Python 2.x (probably 2.7)
• Pyproj
• unzip and zip
• Python library imposm.parser
• Python argh
• Bash
• What pymapmatch needs and argh for argument parsing. Pymapmatch is included as a submodule.

For the lazy, all needed dependencies can be installed in a Debian-like distribution (tested on ubuntu 14.04) using:

sudo apt-get install make swig g++ python-dev libreadosm-dev \
	libboost-graph-dev libproj-dev libgoogle-perftools-dev \
	osmctools unzip zip python-imposm-parser python-pyproj \
	python-argh

Then fetch the sources:

git clone --recursive https://github.com/tru-hy/gtfs_shape_mapfit
cd gtfs_shape_mapfit

and build the required binary stuff:

make -C pymapmatch

You'll need a OSM export in XML or PBF format for the area covering the wanted shapes, referred in examples as map.osm.pbf. PBF is a lot faster, so use it if you can. You can download a suitable area extract from http://download.geofabrik.de/. As an example, for finland this is:

wget http://download.geofabrik.de/europe/finland-latest.osm.pbf -O /tmp/finland-latest.osm.pbf

To reduce memory and computing time significantly, the map should be cropped to only the area needed. There's a script shapes_bbox.py for getting the suitable bounding box. This can be used with eg. osmconvert to clip the map. Assuming you have osmconvert and a GTFS zip file as /tmp/google_transit.zip, run:

osmconvert /tmp/finland-latest.osm.pbf -b=`./shapes_bbox.py /tmp/google_transit.zip` -o=/tmp/map.osm.pbf

In addition to the map file and GTFS zip file, you'll need to specify the map projection used during fitting. This is specified as a PROJ.4 string. The default parameters assume the output is in meters on a 2D plane. Using a projection not suitable for your geographic area may cause very bad results.

For example in Finland a good choice is the ETRS-TM35FIN projection, which has the EPSG number 3067. As a PROJ.4 string this is +init=epsg:3067.

With the map file, GTFS zip and the projection string, the shapes and stop locations can be fitted on the map data with the fit_gtfs.bash script. Using the values discussed in above sections and resulting to a new GTFS zip file /tmp/google_transit.fitted.zip, this would be:

./fit_gtfs.bash /tmp/map.osm.pbf +init=epsg:3067 /tmp/google_transit.zip /tmp/google_transit.fitted.zip

This will take some time, depending on your hardware, map size and number of routes. Helsinki region with about 1800 routes takes about 40 minutes on a Intel(R) Core(TM) i3-2310M CPU @ 2.10GHz. The performance should scale almost linearly with number of cores.

Currently only buslines, trams, subways and trains are fitted.

If any of the measurements aren't in the search radius (default 100m) away from the right road, the results can be very bad. Also if roads are marked wrong in the used map (eg. one-way street where it's actually two way), very weird errors may occur. The script tries to detect bad errors and uses the original data instead in such cases. These are also printed as output log of the fit_gtfs.sh command.

The performance is reasonable, but could be quite easily made a magnitude or three faster with a neglible chance of non-optimal fits.

Probably many more.