High precision mapping with streetlevel imagery

Friday 2:40pm, Main hall

Contribution & data collection

(2a) Data users: Commercial

Oindrila Gupta 30min


Mapillary and OpenStreetCam are crowdsourcing nearly 1 million kilometers of streetlevel imagery per year. This imagery allows detection of highly detailed map features like points of interest, amenities, traffic signs, street names, building levels names and number - all impossible using only satellite imagery.

This talk will focus on how we use streetlevel imagery at Mapbox every day to create a high precision map dataset, that is the closest to reality. We map navigational features like turn restrictionss and points of interest, verify user feedback, and even detect harmful or incorrect changes to the map. We will talk about how we built the OpenStreetMap Navigation Map (https://mapbox.github.io/osm-navigation-map/) using Mapillary’s traffic sign detection API, and how anyone can use it to map navigational assets and the massive future potential for validating edits with streetlevel imagery.