Mapillary uses Amazon Rekognition to work towards building parking solutions for US cities

Mapillary is a collaborative street-level imagery platform that allows people and organizations to upload geo-tagged photos, which can then be used by customers to improve their mapping systems or applications.

Mapillary uses Amazon Rekognition, a deep learning-based image and video analysis service, to enhance their metadata extraction. By using the DetectText operation from Amazon Rekognition, Mapillary is able to detect and extract text from images of traffic and parking signs to enrich mapping data.

“We’ve found the accuracy for detecting text with Amazon Rekognition to be both consistent and accurate,” says Yubin Kuang, Computer Vision Lead at Mapillary. “Mapillary has already built some of the world’s best-performing computer vision for street level-imagery, so integrating the DetectText API from Amazon Rekognition means we can extract text from traffic and parking signs that are automatically detected by Mapillary technology. This workflow shortens our developments cycles and allows us to get an overview of parking infrastructures at scale.”

Architecture

Mapillary stores hundreds of millions of images in Amazon S3. These images are uploaded to Mapillary by people and organizations everywhere. By using the DetectText API operation from Amazon Rekognition within Mapillary’s traffic sign detection pipeline, the metadata extracted from street-level images are then made searchable by using an Elasticsearch cluster. To illustrate, the following example shows an image uploaded by a user to Mapillary. The green box highlights the traffic sign detected by Mapillary, with the white boxes highlighting the text that Rekognition detects, extracts, and converts into machine-readable text.

“This work with Amazon Rekognition allows us create a solution to help improve parking across US cities,” says Jan Erik Solem, CEO and co-founder of Mapillary. “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. We believe that working with Amazon Web Services, and Amazon Rekognition in particular, will help us work towards creating automated, computer vision-driven parking solutions that help cities save time and money.”

About the Author

Kaiser Larsen is a Product Marketing Manager for AWS artificial intelligence solutions. Outside of work you’ll find him hiking, cooking for family and friends, and eating ice cream whenever there’s an excuse to celebrate.