Mapillary Summary

Mapillary is a startup founded in Sweden in 2013 by Jan Erik Solem, the founder of the facial recognition company, Polar Rose that was eventually acquired by Apple (Lunden, 2016).  The vision behind Mapillary was to create a more open and fluid version of Google’s Street view.  To remedy the limitations of a solo team with a camera rigged to a vehicle, Mapillary created an open platform utilizing the concept of crowdsourcing to build a better, more accurate, and personal map.  In addition to creating a better map, the company analyzes the photo to collect geospatial data.   Mapilliary uses a technology called Structure from Motion to create and reconstruct places in 3D.  By using semantic segmentation on the images, the company seeks to understand what is in the image such as buildings, pedestrians, cars, etc and build that into AI systems.  As of May 2017, their database held over 130 million images all through crowdsourcing that is also being used to train automotive AI systems (Lunden, 2017).

 

Mapillary is determined not to use advertising, but instead will focus on B2B platform to provide information for governments, business, and researchers.  Though researchers may use the data at no charge, commercial entities can purchase Mapillary’s services from $200 or $1000 a month dependent on the amount of data used.  The site list several institutions that have successfully used Mapillary.  The World Bank Transportation and ICT group utilized Mapillary to capture images for a rural accessibility project to evaluate the environment and road conditions remotely.  Westchester County in New York state has used the service to capture their trails to create interactive hikes with their park systems.

To date has mapped over 3 million kilometers of over 170 million images on all seven continents.

 

 

To explore Mapillary:

 

  1. Go to mapillary.com
  2. Create an account by clicking on the “Create Account” button on the lower left side of the page.
  3. Choose to create a Mapillary login or use either Google, Facebook, or OpenStreetMap login.
  4. Once signed in you may explore maps or create maps.
    1. To explore maps:
      1. Zoom in on an area on the map-a green line indicates that area has been mapped.
      2. Go to the magnifying glass on the upper left side of the screen and enter a location.
  • When you have located your area, place your curser on the line and a photo will pop up and the image will locate on the lower left screen. Click the forward or back arrows to move through the images or the play arrow.

 

To contribute to Mapillary:

You may upload an image to the webpage:

  1. Click on the menu arrow by your login name on the upper right screen
  2. Click on Uploads
  3. Click on “Upload Images”
  4. Upload your image according to the options and instructions
  5. Click “Review”
  6. You may click on the dot to see the image.
  7. Zoom into the location and place the dot on the map

Or use your smartphone:

  1. Download the Mapillary app on your smartphone from either Google Play or the App Store.
  2. Sign-in or create an account.
  3. Tap the camera icon
  4. Position the camera so that it is level with the horizon and nothing is obstructing view
  5. Choose your capture option: The automatic capture option will automatically capture images as you move every 5 meters OR use the manual option to capture panorams, objects, and intersections
  6. Tap the Red record button and move either by walking, driving, biking, or whichever means of movement and transportation you prefer.
  7. When done, tap on the exit arrow
  8. Tap the upload icon (the cloud icon)
  9. Upload your images-your images will be uploaded and deleted from your device
  10. Images are then processed by Mapillary
  11. You will receive a notification for when images have been uploaded, edits accepted, comments, or mentions
  12. You’re done!!

 

https://techcrunch.com/2016/03/03/mapillary-raises-8m-to-take-on-googles-street-view-with-crowsourced-photos/

https://techcrunch.com/2017/05/03/mapillary-open-sources-25k-street-level-images-to-train-automotive-ai-systems/