firebird854

Hello,

I'm part of the team of three roommates who've been developing a completely free alternative cycling routing service, similar to Strava's route builder/Kommot/RideWithGPS over the past year, https://sherpa-map.com.

Our latest addition is a stunning overlay showcasing what road is what surface type for all of North America, Europe (plus some area around Europe), and Australia from Open Street Map + AI classified data!

You can access this at the bottom of the interface, by simply pressing the "Show Overlay" button.




Black: Paved
Gray: Gravel
Brown/reddish color: Dirt
Tan: Unpaved (misc category that doesn't fit the above)

This can be overlaid on a variety of different map types, including OSM/Google/Terrain/Mapbox, etc.

Not only did it take months of effort to create a pipeline to properly generate this, but the other mentioned source, using AI classification, has been an ongoing research and development effort, that's even prompted us to purchase a whole new workstation in addition to numerous server upgrades (this project has been, well, heating our apartment basement).

This system works by using satellite images of roads that lack a classification, are not primary, or in a town (I've defaulted those to paved), I then get a 640x640 satellite image and use a deeplabv3_resnet101 classifier to segment out/pull be the best 240x240 image of the road out of the image, and use a resenet152 classifier I trained to determine if it is paved/unpaved, then an additional classifier to determine if the unpaved is gravel/dirt/unpaved. It will only accept the classification if it's above a 95% confidence level. I've managed this so far with the American states of Wisconsin/Illinois/Minnesota/Iowa, and will keep on classifying (I've also recompiled our routing software to match)!

I'm planning on implementing a right-click mechanism to allow users to instantly update to correct surface types if they find an issue. I'm also working on adding this data to our very own map tiles, which are in the works.

I aim to keep updating and working on this; if you have any suggestions, or requests for location (I'd be happy to add your region), just let us know in the comments.

Black wallnut 

Where does oiled fit in your surface type? It's not exactly unpaved and it is not exactly paved, although you might think it is paved. It is a base of chip seal overlayed with more chipseal.

Kudos for going forward with your app. I don't see a present need for it where I ride but others might.

Troul 

wonder if this data will be able to be automatically rolled into the "updates" that bicycle computer's push? IE: Wahoo

njkayaker

Bicycle computers typically use Openstreetmap (OSM) data. So, they have access to the OSM road surface data. I don't think the OSM data for that is that great.
One issue is being able to clearly distinguish surface on a small screen.

obrentharris 

I applaud what you are doing but when I look at the overlay in my immediate neighborhood (Point Reyes Station, California) most of the roads have no surface indicated.
Brent

Troul 

just knowing it's not a dirt with the chance of gravel topped surface would be better than assuming the continuing paved road that I am on is expected to remain paved. Fine grey scaled hashing lines to distinguish it not being paved but a continuation of that linear road would be nice.

firebird854

Good question, frankly, I compile a variety of terms into the different "surface" types, as Open Street Map datasets allow users to manually enter any type they desire. Paved = Asphalt, Concrete, Paved, Gravel = Gravel/light Gravel, honestly pretty much every option that contains the word "gravel", Dirt = dirt, Unpaved = everything else, this can range from "Grass" to "Wood" to "Snow" to "Sand" etc.

Also, thanks for your kind words! And hey, I do have two other fun free cycling related sites too:
https://sherpa-map.com/cycling-route-calculator.html --Physics simulator with weather and nutrition data lets you determine how fast you will be with what setup around what mixed surface course.
https://sherpa-map.com/activity-racer.html --Watch in realtime your activities in the same area race each other.

firebird854

This is why I'm using AI to help with this mission. I've just finished building one heck of a work station so I can more easily pull satellite images of roads with no classification and classify them. I've made it through Wisconsin/Illinois/Iowa/Minnesota so far and I intend to cover as much as the world as I can.

You can activate it on the fly by using this tool https://i.imgur.com/ycs7pEe.mp4 but that's using an older model, my new model that I use to augment the overlay is far more sophisticated.

Also, I wrap this data back into my routing engine to help inform cyclists when creating a route.

RB1-luvr

this is neat and easy to use. Thanks.

everything's black in my area :-(

firebird854

Glad I could help! Same situation in my area, Wisconsin... I have to drive 3 hours in any direction to find gravel.

I'm also adding more overlays too, I'm working on adding Strava's heatmap, an overlay that shows popular bike race routes, one that should mtb trails specifically, and one that shows all hiking paths.

mstateglfr 

Interesting project. There is a whole lot of white near me that are paved roads with interstate on/off ramps. I wonder how the tech determines what is/isnt paved since some interstate access roads arent determined. A lot of gravel 15mi from me is white too- unsure what type of surface, I guess?
Most everything is spot on and accurate on where some roads go from paved to gravel.

firebird854

Thanks for the feedback!
White is simply the road color for the default map under the surface overlay, if a road doesn't get a coloring from the overlay, it means there is no data in OpenStreet map and I haven't gotten to that area with my AI classification yet.
If you meant something else with your mention of white roads, please correct me.

prj71

Functionally it seems to work well. However, I laid out a route (I'm in WI also) but it didn't get all of the surfaces correct which would make it a no go for me at this point. Ride with GPS seems to be getting the surfaces correct more accurately at this point.

I've also found this helpful...

Gravel Road Maps and Biking Routes - Gravelmap