After Competition, OpenStreetCam Can Now Detect Dozens Of Sign Types in Australia and New Zealand

Last December and January, OpenStreetCam held a image collection competition in Australia and in New Zealand. The three mappers in each country who collected the most points during the months of December 2018 and January 2019 could each win a gift card: $100 for the winner, and $25 for the second and third place. We just announced the winners to the communities in both countries. Congratulations to steve91, robbie-bloggs, ConsEbt, david-blyth, ivss-xx, and nicknz!

User-uploaded image: Screenshot 2019-02-06 15.50.05.png
OpenStreetCam coverage in Melbourne after the competition

We decided to do these competitions because we wanted more mappers in Australia and New Zealand to get acquainted with OpenStreetCam, and consider contributing to this free and open platform for street-level images. The more contributions, the more help OpenStreetCam can be for OSM mappers! There weren’t many contributions in either country yet, and if you go to the OpenStreetCam web site, you’ll quickly see that there are still large gaps to fill. Still, OpenStreetCam coverage grew by 800% since the beginning of December.

Head over to the ImproveOSM Blog for a step-by-step guide on how to get started with OpenStreetCam yourself!

User-uploaded image: Screenshot 2019-02-06 15.52.07.png
OpenStreetCam now sports almost 1.5 million images in Australia and New Zealand combined

OSM mappers can use the OpenStreetCam images to help with mapping. You can’t see everything from an aerial image. Signs are a great example of useful mapping information that requires an on the ground perspective. This is where OpenStreetCam is particularly handy, because we detect an increasing diversity of signs that appear on the photos for you automatically, using an open source machine learning platform.

Interested in a more in-depth look at the OpenStreetCam sign recognition AI? Have a look at our talk at State of the Map 2018 in Milan or our talk at State of the Map US last fall!

Using detected signs in JOSM


For the sign detection platform to work and detect a variety of signs reliably, it needs training data. Your contributions during this competition have been invaluable to reach that goal. Our Map Team looked at tens of thousands of images collected by the community during the competition in Australia and New Zealand, and validated more than 160,000 traffic signs found in these images. After feeding that data into the platform, we can now reliably detect more than 80 types of signs in Australia and New Zealand. As we continue to look at more images that you contribute, the system will get smarter and we will detect more different types of signs.

Do you want to help train the OpenStreetCam sign recognition AI? You can do this right from the OpenStreetCam web site. Read all about it in this blog post.

Validating automatic sign detections is a quick an easy way in which anyone can help improve OpenStreetCam’s ability to detect signs reliably