Mapillary crowdsourcing adds new features to maps
Mapillary, founded in 2013 to make “street-level imagery and map data available to everyone,” has announced that it can now extract 42 new types of map features from street-level imagery and automatically position them on a map. In a company blog post at blog.mapillary.com, Gerhard Neuhold, Computer Vision Engineer, wrote that the features include utility poles, streetlights, mailboxes, and manhole covers. This will help cities, mapping companies, and transportation agencies keep their maps up to date using cameras.
The company brings together a global network of contributors with the goal of making map features accessible to everyone by visualizing the world and building better maps. Anyone can join and collect street-level images, using simple tools like smartphones and cameras, and the company’s computer vision technology can connect images to create immersive street-level views and extract map data. Mapillary has grown into a worldwide community, with people and organizations having mapped more than 5 million km across 190 countries.
“Mapillary is on the mission to help fix the world’s maps by making map updates available at scale,” wrote Neuhold. “The old way of collecting data and editing maps manually is too time-consuming to keep up with how fast cities and roads are changing. This is a growing problem; mapping companies, transportation agencies, and city governments alike are struggling to keep their map data up to date.”
The company has been able to detect about 1500 different types of traffic signs for some time already, resulting in 18 million traffic signs being automatically placed on the global map. The 42 new object classes are applicable to use cases such as crosswalks, street lights, and benches for pedestrian mobility; bike racks and bicycle traffic lights for cycling; lane markings, traffic lights, and parking meters for transportation; manholes, utility poles, and trash cans for public works; and traffic cones and construction barriers for road maintenance projects.
These add to the company’s full list of supported object classes, more than 186 million objects as map features across the 430+ million images that have been contributed to the Mapillary platform from all over the world. To get started, companies wanting to check out the new features can retrieve data files or via the API by getting a subscription through Mapillary for Organizations.
“With map features, the quality of the outcome depends on both the technology as well as the input—that is, the captured imagery,” wrote Neuhold. “Shortly put, we use computer vision to detect objects in images and reconstruct places in 3D. By combining the two, we can estimate the coordinates of each object, and make that data available as map features.”
To estimate the location of an object on the map, the object needs to be detected in two or more images. The positions of the images are then used to calculate the position of the map feature. The accuracy of the location of the map feature is influenced by the accuracy of the location of the images, so the company encourages using a high-precision GPS device when capturing. More images in the area mean more data points to triangulate the position of the object more closely, so accuracy is improved by capturing images more densely.
The new object classes are available in beta, under the same subscription plan as traffic signs. On the technology side, the company is focusing its efforts on making the detection algorithms more accurate and improving the quality of our 3D reconstruction.