These scripts can be used to inspect matchings generated by the map matching plugin for OSRM.

There are two ways to install the server side component. One involves building osrm-backend and node-osrm from source and one needs binary packges of a special node-osrm version.

npm install osrm

Which downloads a binary package if it is available on your platform.

Compile OSRM with debug support:

git clone https://github.com/Project-OSRM/osrm-backend.git
cd osrm-backend
mkdir -p build
cd build
cmake -DENABLE_JSON_LOGGING=1 .. 
make && sudo make install

Run:

npm install osrm --build-from-source

To download and compile node-osrm from source.

Run:

bower install && make

To install the front-end components.

To import traces in GPX or CSV format contained in a folder data to the labeling database run:

node bin/server.js data

This will create a file data/clasification_db.sqlite which will contain a list of all traces and their classification.

Assuming your GPX traces are contained in a folder data in the current repository root:

Locally run:

node bin/server.js data path/to/dataset.osrm

Alternatively if you want to use osrm-routed instead of node-osrm just run:

node bin/server.js data

Which expects a osrm-routed server listening on https://127.0.0.1:5000.

Now you can view the frontend on https://127.0.0.1:8337 in your browser. It will look somewhat like this:

Which shows an interactive trellis diagram of the matching. Select a state pair to view the transition probabilities and Viterbi values.

You can use the left and right arrow key to cycle through the traces.

Opening https://127.0.0.1:8337/classify.html will display a minimal interface for easy classification. Pressing 0 will classify as unknown, 1 as valid and 2 as invalid.

The labels will be saved in classification_db.sqlite which can be used by bin/test_classification.js to verify the classifier implemented inside the OSRM plugin. bin/test_classification.js will also generate tested_db.json which is needed by bin/calibrate_classification.py to derive better classification values.

You can batch match a dataset using bin/traces2geojson.js data > matched.geojson which produces a geojson file containg the following features:

For every sub-matchings:

{
        "type": "Feature",
        "geometry": {
            "type": "LineString",
            "coordinates": [[lon, lat], ...],
        },
        "properties": {
            "type": "matching"
            file: "path/to/file"
            confidence: 0.5 # in [0, 1] -> 1 means very confident, 0 means no confidence
        }
}

For every trace fragment:

{
        "type": "Feature",
        "geometry": {
            "type": "LineString",
            "coordinates": [[lon, lat], ...],
        },
        "properties": {
            "type": "trace"
            file: "path/to/file"
            confidence: 0.5 # confidence of the corresponding sub-trace
        }
}