The following is work done at Lincoln Laboratory, where we made 3D reconstructions of city-cized models.

3D Structure from Motion

Many of the tools have now been automated by ChangChang Wu’s Visual SfM software package. There are a few installation tricks, but for the most part, this program makes it incredibly easy for a non-photogammetrist to build models without having to write an inordinate amount of code. For example, meet Rosie, my dog.

She was sleeping one day (much like she was in the above picture), and I took a few pictures of her over several different angles, and then ran SfM algorithms to make a 3D rendering:

The advantage of these methods is the fact that they can be used with images in the wild (for example, scrape FlickR, YouTube, or whatever), and organize them by using feature extractors and matching those features with each other. So, if run around . Below is a reconstruction that we built from 2317 images just running around the Stata Center near MIT campus.

Once you’ve reconstructed the city with images, you can align that with other interesting 3D point clouds from other modalities. For example, you can take a LiDAR point cloud, align it with the picture point cloud, and voila! You get this:

If you’ve done everything right, you should be able take any new photo, and then align that new photo with the point cloud to figure out where you are!