Mapping the world with photos is the vision of Mapillary, but what makes the start-up stand out is its mission to fix maps by making them more detailed through the use of computer vision.
To further develop their platform, Mapillary recently announced working with Amazon Rekognition to detect and read text in Mapillary’s database of 350 million images.
With an objective to address the parking woes across US cities, the collaboration means that Mapillary’s technology detects parking signs in images on its platform, before running Amazon’s Text-in-Image Rekognition to extract text from the parking signs. The process allows Mapillary to extract parking sign data from the 350 million images it has on its platform, all uploaded by tens of thousands of users across 190 countries. The initiative means that cities everywhere can get an automated and computer vision-driven overview of parking signs and data.
Advent of Mapillary
Having always been interested in building technology that solves real problems, Mapillary Co-Founder and CEO Jan Erik Solem says, “I realized that even the biggest mapping companies are having trouble getting all of the data they need to complete their maps. There is a lot of map data out there, but no one was sharing it.” Also, there are places everywhere that don’t get covered by traditional map providers. Allowing people to map places that matter to them and then making the map data available to everyone is at the core of what Mapillary does.
Mapping the world
The process of updating maps has traditionally been slow and expensive. Gathering street-level imagery has involved sending out fleets of cars equipped with sensors and cameras, and this has predominantly been done in areas where there is a commercial interest in mapping. Then once the imagery is collected, it is analyzed manually in a labor intensive process.
A company that allows anyone, anywhere to capture and upload street-level imagery, using any device is Mapillary. By using the Mapillary mobile app you can take pictures with your smartphone, or capture images with an action camera and use upload tools. All of the images are automatically analyzed using computer vision to detect map data. Using computer vision for map data detection is much faster and more accurate than having humans manually analyze street-level images. The entire process makes Mapillary the single most scalable way to keep maps updated.
Since anyone can use Mapillary, Solem reiterates that any place on the planet can now be mapped from the perspective of any one group — including places that were previously untouched or inaccessible to mapping fleets. “For instance, Cubans mapped Havana using Mapillary. The citizen advocacy group Bike Ottawa, meanwhile, used Mapillary to map Ottawa from a cyclist perspective in order to highlight cyclists’ safety concerns across the city,” says Solem.
Solving problems for many
Mapillary benefits different segments in different ways. Solem says, “We have customers in the mapping, GIS, and automotive fields. Cities and towns need to do street asset inventories on a regular basis and use Mapillary rather than spending hundreds of thousands of dollars on contractors, or putting the time into visiting every street in their area on foot.” Since the platform automatically extracts map data, the company allows for quick and convenient inventories of things like traffic signs. The City of Amsterdam, for instance, uploaded imagery covering all of Amsterdam and could within a week access map data from every corner of the city.
Large parts of the automotive sector are right now focusing on developing autonomous systems for driverless cars. These systems need highly detailed and up-to-date maps. They also need to be trained on masses of data in order to learn how to see and navigate in street-level environments. That’s why the company has built the Mapillary Vistas Dataset, the world’s largest and most diverse training dataset for street-level image segmentation that is publicly available. The Vistas Dataset is used by the likes of the Volkswagen Group.
Sources used to collect data
Mapillary is collaborative, meaning anyone, anywhere, can contribute and use the data. The dataset is growing at an exponential rate. Solem elaborates, “In August this year we passed 350 million images on the platform. Around 80% of the images come from individual contributors. We also have cities, Departments of Transportation, and other organizations upload their imagery to Mapillary as it helps them fix their own maps and get an understanding of their assets.” The Department of Transportation in Arizona, for instance, recently uploaded 4.7 million images covering all of their highways. People and organizations contribute imagery for different reasons — but once the imagery is uploaded to Mapillary, it’s made public so that anyone can access the imagery and the data in that imagery.