LCSSMatcher produces terrible results for complex highway interchanges

[jnewb1]

highway.csv

Running this highway gps trace through LCSS matcher following the lcss-example.ipynb notebook gives terrible results around the interchanges.

Red is the matches, blue is the trace

[nreinicke]

Sorry you're having issues. This is likely caused by having a geofence does not properly capture the road network you want to match to. For example, with the default geofence boundary of 2000 meters from the example, the road network that gets downloaded is missing some links in that specific area:

To fix, you could try increasing the boundary of the geofence (I tested with 5e3 meters: geofence = Geofence.from_trace(trace, padding=5e3)).

I also noticed that there are some match errors at the start and the end of the trace. Those are unexpected and we can take a closer look but increasing the boundary should get you better results.

[jnewb1]

Ah yes, that seems to have fixed it. Thank you. Just curious why the geofence of 2km doesn't match any points on this part of the road? I feel like having to have a 5km buffer seems a little extreme when the points are right on top of the highway

[nreinicke]

Just curious why the geofence of 2km doesn't match any points on this part of the road?

That's a good question and I can't give a concrete answer without looking at it in detail but mappymatch takes the largest strongly connected component from the graph that gets returned by the geofence and so it's likely that there is some graph connectivity issues towards the edge that are causing those segments to be excluded from the main SCC.

[jnewb1]

Here's another example...

highway2.csv

5km geofence

No road for this part?

Terrible match :(

At 6km buffer then it starts working...

so it's likely that there is some graph connectivity issues towards the edge that are causing those segments to be excluded from the main SCC.

strange thing is that even disabling all filtering in osmnx still doesn't return this portion of the highway at 5km buffer, but then starts working at 6km...

[nreinicke]

Yeah I looked a little deeper and can isolate the problem to when we extract the largest strongly connected component.

Looking at the raw osmnx graph for the 5km buffer we see this:

Then, after extracting the largest SCC we see this:

If we expand the buffer to 7km we see these graphs before and after extracting the largest SCC respectively:

Right now we rely on extracting this largest SCC to make sure we can cleanly route through the network but there are certainly more sophisticated ways to handle this that would allow for smaller buffers. But, for now the solution would be to use a generous buffer and make sure that all of the roads are present.

[jnewb1]

But also increasing to 10km buffer then we get no matches, even though map seems fine

Which means I couldn't just leave it at a high buffer for all cases. :(

[nreinicke]

Interesting, it looks like you've uncovered a bug where we don't properly handle cases where there is only a single road to match to as in this case. I've applied a fix in b351b31 and it should be fixed in the next release.