Amsterdam in 360: From Imagery to Map Data in Seven Days

The City of Amsterdam captured 800,000 panoramic images of the city and brought them to Mapillary in an open data effort. In just a week's time, the street-level views and the map data derived from the imagery have been made available for anyone to use.

The City of Amsterdam captured 800,000 panoramic images of the city and brought them to Mapillary as an open data initiative. In just a week's time, the street-level views and the map data derived from the imagery have been made available for anyone to use.

Amsterdam map

The City of Amsterdam has collected over 800,000 panoramic images using the powerful Trimble MX7, which until recently were only available on the community’s open data portal called City Data. Last week, however, they decided to upload this imagery to Mapillary. Starting on June 21, we began to process the imagery to be fully interactive on Mapillary’s worldwide map. During the week’s period since then, we’ve also derived a multitude of data from the imagery with the help of computer vision technology.

The imagery captured by the city is stunning in and of itself. Not only are roads captured but additionally, the 360-degree camera was attached to boats in order to show the canals (which Maptime Amsterdam mapped previously).

Amsterdam is a beautiful city replete with remarkable architecture, a seemingly endless supply of bridges, a famous legion of bicycles, and of course, the charming labyrinth of canals. While the imagery shows off all of this, we are now able to take it a step further—not only does our technology see all these objects just as you do, but it also relates them to the geographic position of the image. While many of us could take several weeks, months, or even years to explore and document the entire city, a multitude of objects in thousands of images have now been detected over the course of a week.

The resulting database showcases the locations of more than 80 classes of objects throughout Amsterdam, as visible in the web map above. To get started, try activating the benches layer and see if you can spot a bench in the photos on the map! Make sure to click and drag on the image viewer to get the full 360-degree experience. Click any of the buttons for the included example classes to load them on the map. Click any point location to view the corresponding 360-degree image containing the detected object.

There are many parts of the derived data that are worth special note. For a simple example, below we can see how closely the water classification in Mapillary images matches the location of Amsterdam’s canals.

Amsterdam canals

We also can see how the tunnel classification sits precisely on top of the IJ-tunnel to the north of the city center.

Amsterdam tunnel

Tunnels are very rare on Mapillary, as everyday GPS devices typically struggle to record location underground. The Amsterdam images display in the tunnel due to the high-grade GPS and sensors, which are not built into most smartphones. This allows Mapillary to detect and map the location of this tunnel, among others, aiding in a speedy inventory of city infrastructure.

Also notice that the images with bridges in them are often located directly adjacent to a bridge, as seen below, showing that this is a great way to inventory another type of infrastructure.

Amsterdam bridges

Lastly, we can get a good look at the locations of bicycles in the 360-degree images. While this doesn’t tell us the distribution of bicycles across both time and space, it can start to give us an idea of where bicycles can most often be seen in the city (well, besides everywhere!). Using timestamps written into this same dataset, we could also start classifying densities by day of the week or time of day.

Amsterdam bicycles

Overall, datasets like these are unique in the world of GIS. They are generated by processing geotagged images with computer vision in order to classify every pixel in the image and give it an approximate geographic location. While all the cases in the Amsterdam imagery show the location of the image containing the detected object—and not the location of the object itself—it is also possible to pinpoint these objects in space as we currently do with traffic signs. For example, the map below shows the location of all detected traffic signs in Amsterdam, rather than the images in which they were detected.

Producing map data from 360-degree imagery is possible thanks to a thorough image capture project across the entire city of Amsterdam. While collecting images in such a large urban area can take some time, the map data output can be very swiftly generated. The imagery was made available on Mapillary from June 21st and just seven days later we have a plethora of resulting data available. 800,000 photos of Amsterdam have proved to reveal more than meets the eye and in the short span of one week contributed to a better understanding of the city. Explore the full web map here and feel free to contact us with questions about this project!


Continue the conversation