Tinkering With Creating 3d Point Clouds From Video Imagery

The word "hyperlapse" has been getting a lot of play lately given Instagram's decision to incorporate the ability to create stabilized, depth-masked, time lapse video directly in the application. Looking into how it was done, I ended up stumbling on the detailed and technical Microsoft video which was deeply inspirational to me. But I had to go one further – I had to step beyond something as simple as basic hyperlapse.

I had to move into the field of reconstructing entire spaces and remapping them, provided only images or video (or both).

Microsoft had already plowed that fertile field with a project they referred to as Photosynth (https://photosynth.net/preview/users/SquidLord), which basically just does for a series of photos what the new "hyperlapse" does for video time lapse – pretty much exactly. Underlying all this technology is the ability to generate a 3-D point cloud which describes the space/object being photographed, and the ability to generate a mesh of polygons from that data, taking color information and texture information from the photographs themselves projected back onto that mesh, to turn up a 3-D, explorable model of the original target.

Interestingly, you can do all of this with publicly accessible and free software.

Because I'm a geek, this is exactly the sort of project I want to apply myself to.

Here's a first pass.

Software used:

VisualSFM: http://ccwu.me/vsfm/

Introductory Guide to VSFM: http://www.iafsm.org/resources/Opensourcetoolspart3.pdf

Meshlab: http://meshlab.sourceforge.net/

Help us caption & translate this video!

http://amara.org/v/FRo2/ ... https://www.youtube.com/watch?v=TU2hoN3EQzg