Follow Datanami:
November 19, 2015

Mapping Startup Leverages Machine Vision

A Swedish mapping startup that uses machine vision to stitch together street-level photos is ramping up a service that uses proprietary algorithms to connect crowd-sourced photos of street scenes “over space and time.”

Mapillary, Malmö, Sweden, said its app can be downloaded to a smartphone to take pictures, for example, of street-level scenes used to create maps. Once photos are uploaded, the app combines them with other photos to create what amounts to a 3-D map of a route.

As of this week, Mapillary said it has stitched together more than 42.6 million photos, creating more than 1 billion meters of navigation routes.

Along with apps for iOS, Android and Windows smartphones, the mapping service also includes a set of APIs for using photos and data extracted from images.

Using machine vision to extract information from images, the service “automatically match[es] features to the ones in photos from the same geographic vicinity so that we can compute how the images relate to each other and how to navigate from one to the other,” the company explained in a recent blog post.

Going a step further, the mapping service also can recover a 3-D model of the area photographed. Leveraging that data results in the creation of immersive 3-D views. The startup claims its algorithms can detect traffic signs and blurred faces as well as allowing users to “navigate photos” and virtually explore landscapes in 3-D.

In order to improve the mapping capabilities of its computer vision approach, Mapillary said earlier this month it has started using more detailed mapping shapes representing countries, regions, localities and neighborhoods. It now has access to 233,000 map shapes based on a project by partner Mapzen, an open-source “mapping lab,” to maintain the outlines and hierarchies of shapes used on maps. Mapillary said it previously relied on only about 12,000 shapes that lacked the detailed of new ones.

The idealistic startup notes that current mapping alternatives are limited to images of streets taken from cars equipped with camera rigs. That approach leaves out places without streets or views that are not updated frequently. Hence, Mapillary relies on crowd-sourced photos from around the world to map isolated areas.

“Don’t take tens of photos, take hundreds or thousands at a time. The more the better. Our servers can handle it,” the startup declares in a “manifesto” on its web site.

The uploaded photos are combined into street-level views and those views are stitched together into country and regional maps that users can drill down to the level of individual neighborhoods and businesses.

The startup’s API helps build applications and integrations using Mapillary’s database of photos and related image data. It also offers developers “data tiles” used, for example, to display photo data on vector maps. In addition, a JavaScript widget allows developers to embed a viewer using an iframe tag.

Datanami