Mapillary and Amazon Rekognition collaborate to build a parking solution for US...

Mapillary and Amazon Rekognition collaborate to build a parking solution for US cities through computer vision


US: Mapillary, the street level-imagery platform that uses computer vision to fix the world’s maps, has started a collaboration with Amazon Rekognition to end 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.

The collaboration taps into a pressing issue, with research from 2017 showing that congestion, parking tickets, and time spent looking for a parking spot cost Americans more than $73 billion annually. Meanwhile, city authorities across the US are grappling to update inventories of parking signs and data, making it difficult for city planners to maintain and plan their parking infrastructure. Mapillary’s approach means cities are now able to access parking sign data at scale, with the hope that it will improve parking infrastructure.

Jan Erik Solem, CEO and co-founder of Mapillary, says: “City authorities are struggling to keep track of their parking signs and data. For cities to manually track this themselves would cost millions of taxpayers’ dollars, not to mention the enormous time investment. The collaboration between Mapillary and Amazon Rekognition means that several cities in the US are now in a position to get complete updates on their parking signs through an automated and computer vision-driven process, saving both time and money.”

As part of the initial tests, Amazon’s Text-in-Image technology was run across more than 40 000 parking signs that were automatically detected in street-level images from the Mapillary database. The results showed that 95% of the signs were nearly perfectly decoded, meaning that 95% of readable text lines were correctly detected, with only very few small or blurry characters not recognized correctly. The remaining 5% have missing pieces of text or minor errors in decoding.

Mapillary’s parking sign detection is likely to be particularly interesting to cities, as misinformed perceptions of parking space cost city authorities millions of dollars. Meanwhile, the rise of ride-sharing apps and semi-autonomous vehicles is redrawing the map of urban mobility, putting increased pressure on city authorities across the United States to get a better understanding of their parking infrastructure.