Mapillary coverage is growing fast and the collected imagery is great for mapping turn restrictions and speed limits in OpenStreetMap. Here’s how

The Mapillary team is moving fast, the recently added street sign detection feature is highly useful. Here’s my wishlist for an even smoother workflow

  • Better mapillary support in JOSM
  • Fine tuning of the blurring algorithm to not exclude street signs like here
  • Larger mapillary image in iD
  • A link from an image on to that opens iD with the right image present

I spent some time working with Mapillary. I chose three routes from a variety and looked through them. In contrast to the panoramas, such as Google and Yandex (Yandex has given the permission to the usage photos from their panoramas in OpenStreetMap), Mapillary has several limitations. The imagery of Mapillary is basically not panorama, because images are taken in the one direction. We should be very careful to identify all the necessary details (there are also pictures out of focus).

out of focus

I tried to answer to some questions and I can say next:

Can we use Mapillary for mapping turn restrictions?

Yes, we can use Mapillary for this purpose in combination with satellite images. Since Mapillary shots are not always taken in the right place and it is just addition to satellite images. I tried to add to the OpenStreetMap rules of turn lane restrictions at intersections. There are two schemes for this in OpenStreetMap (scheme 1, scheme 2). I used scheme 2 as it has a convenient plug-in in JOSM.

It is necessary to determine which scheme is preferable.

Can we use Mapillary for mapping max speed?

Yes, speed limit signs can be discerned on images, but it would be nice to know the local traffic features or rules. max speed

How is the editing workflow with OpenStreetMap, what could be better?

I’ve worked with Mapillary and JOSM since JOSM is more familiar for me. I know that Mapillary layer is present in the iD, but the picture, that is displayed in the iD, is too small to see the details. I was moving between Mapillary shots on the website and, if necessary, using the control elements on the site to enlarge the picture to see the details. In the case of finding objects that can be added to the OpenStreetMap, I used the link under the image for editing in JOSM. For more accurate orientation I used Mapbox satellite images as a background on the Mapillary site and JOSM. Sometimes I had to spend a lot of time in order to establish the position of objects on images. It would be easier if the pictures could be enlarged in iD. It would be great if the same place from Mapillary could be opened in ID with the same photo in the preview.

But back to Mapillary.

  • One inconvenient thing is that if you change the zoom of the captured image the control of navigation between images disappears.
  • In the case of car going at high speed, many shots are out of focus and there can be seen only general details on them.
  • The algorithm of blurring signs and people faces in Mapillary should be reviewed, so that the inscription on road signs were not processed.

    blur of sign

  • And one more wish to Mapillary users is to install the camera at the angle to the direction of motion, rather than the direction of travel. Thus, we could descry more details which are on the side of the road.

What other information could we gain from Mapillary?

Except of turn restrictions and speed limitations we can add information about all objects which we can discern in the picture and which we are interested in. Among them: the number of lanes on the road, allowed turns from a given (turn) lane, the presence of cycling lanes, sidewalks, crosswalks, traffic lights, possible type of road surface; as well all kinds of POI, banks, shops, fire hydrants, petrol stations, cafes, and information contained on road signs.

P.S. After this article was prepared, Mapillary has introduced a new feature of Traffic Sign Recognition on images.

All images by Mapillary / CC BY-SA 4.0