Train and tram tracks are dangerous for cyclists - How can we avoid routes involving these lines?

[tsobuskerudbyen]

We have data now, from both Sweden and Norway, that train and tram tracks are over-represented in accident data sets (Norway: https://www.toi.no/cywalk/ )

Ticket is about:

1. What is accounted for this in OTP today?
2. What can be done to warn about and route cyclists away from parallel railway lines? (With e.g. existing or new road tags and/or proximity data as it may be that in general tracks are drawn separately)?
3. How much will it cost to implement?

Example how it is tagged today
Oslo tram tracks as part of the road: https://www.openstreetmap.org/way/4649716 - Google Street View: https://maps.app.goo.gl/deXbcDrZWJ4E9JedA
Kristiansand train tracks as part of the road: https://www.openstreetmap.org/way/376508066 - Google Street View: https://maps.app.goo.gl/9yiVH9Dqyzgx2rwP8

[leonardehrenfried]

I investigated this today and can say the following:

1. It's not accounted for at the moment at all.

The rest of the questions are a bit more difficult to answer. In the OSM links that you provided and with a look at Oslo tram lines in OSM I can see that they are mapped mostly as separate ways, which the OSM wiki recommends.

However, I also see that in a few places the roads are mapped as embedded_rails=tram, which is the correct tag usage for the situations we are talking about here. The place in Oslo that you linked is not tagged that way. Do you know why?

Now, lets say that all ways which have embedded tram tracks are tagged correctly, what would you like to happen?

1. OTP could give them a penalty, which means that they are avoided. How much they are avoided can be calibrated.
2. A message for the UI could be added to the routing result.

If all ways are correctly tagged then 1. is easy to do. 2. is harder and needs a bigger budget depending on what the exact requirements are.

Lastly, if you we need to have some proximity-based algorithm, the budget would need to be quite high. We have a few of those and they are very difficult to get right. One example was the application of the street name to a separately-mapped sidewalk. Even after a few weeks of tweaking the success of proximity-based algorithm was moderate.