The Boulder 3D Experiment

The Event

On October 6th, 2019 twenty intrepid photo mappers showed up at the Bohemian Biergarten to see if we could collectively create a 3D model of Downtown Boulder. Together we generated one hundred and seventy one 3D models from roughly five thousand photos. In our biased opinion the results were pretty incredible in both quality and scale for a few hours work.

Sample of Pixel8 Mobile App Models from the Boulder Event
All 171 Collects of Downtown Boulder

Sharing the Data

Now comes the fun part of sharing the data back to the community of both local mappers and the broader mapping and 3D collective. We’ve posted interactive maps of the 3D models and the ground control surface at Pixel8.earth. In addition there are links to the point cloud data in both .ept and 3D tiles formats. The data/imagery is licensed as Creative Commons Attribution 2.0 Generic. Long term we’ll implement a dual licensing approach to make the effort sustainable. In the short term we want to make it as open as possible. Our hope is the community will take the data and remix it with other imagery and 3D data sources. Take the point clouds and generate cool meshes. Do things we’ve never thought of. We’ll be experimenting and sharing results in parallel.

This is all driven by the team’s belief that the future 3D map of the globe will be a hybrid of many data sources. No single imagery source or sensor is good enough to cover all the use cases. It sounds cheesy but we are literally better together. To this end we want to help facilitate experiments that drives data source conflation and alignment forward.

Lessons Learned

One element to this puzzle is enabling commodity photo with EXIF metadata to be part of the solution. To this end the Boulder experiment taught us valuable lessons for applying crowdsourcing. A few takeaways from our first test of the Pixel8 mobile app in the wild:

  • There is a good amount of variability in the quality of photos collected from first time user in regard to creating good photogrammetric models.
  • We created a “best practices” guide, but having a training session virtually before the mapping event or in person before everyone heads out would help a lot.
  • Creating more interactive feedback “in app” to guide users to take better photos would help as well (e.g. show photo overlap as pictures are taken).
  • Allow multiple users photos to be pooled to create a single photogrammetric model for a discrete location vs. solely stitching models together post generation.
  • Getting data back to the community more quickly. It was a great stress test and surfaced lots of challenges, but we need to turn data around faster than two weeks. The event helped enable the changes we needed to make to improve.
  • Bundle adjustment with large graphs is computationally intensive but super effective. We’ll need to work on optimization to include it in the next pipeline.

While we definitely found lots of areas we can grow and improve; we were also psyched that the core premise held. A small group of volunteers could map a large percentage of Downtown Boulder in a few hours with just photos. In our mind this really opens up a new universe of mapping tools at a time when creating hyper local 3D raster maps will be important. That said it is only one piece of the puzzle and while we nail down improvements we’ll also be expanding the realm of imagery sources we can tap to grow the scale and accuracy of the collective 3D map. Here is a quick preview adding Nearmap’s impressive aerial derived 3D point clouds.

Nearmap Aerial Plus Pixel8 Mobile

(Caveat — we’ve reduced the pixel size for the Nearmap data to highlight alignment of the Pixel8 data. Nearmap looks very nice on its own with larger pixels.)

We’ll also be looking to host more photo mapping parties. We have teams in Denver and Austin, but if other locations are interested reach out and we’ll see how we can best support. Long term we’ll have dedicated infrastructure in place, but till then we’ll learn by doing. Please reach out to us with feedback or questions about the data/imagery — sean.gorman@pixel8.earth, @pixel8earth or @seangorman. Lastly many thanks to all the volunteers who came out to help with the experiment. It wouldn’t have happened without you!

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Pixel8Earth

Pixel8Earth

170 Followers

We are building a multi-source 3D map of the globe one image at a time.