Time slice (bullet time) processing into 3D models

As you can probably tell, I’ve been obsessed with aerial 3D modeling lately. I finally saw Inferno, the short time slice / bullet time video, and thought it was pretty amazing.

[vimeo 132024990 w=540 h=304]

After seeing the video, I thought it might be fun to run the frames through Pix4DMapper to see whether a 3D model and camera information could be pulled out of the frozen moment in time. Inferno used 48 DSLRs in an array to achieve bullet time, and in the Pix4D reconstruction, you can see exactly where they all were located.

Pix4D was unable to reconstruct too much of the guy who was spewing out the flames, even when I resized the video frames to be much larger in resolution (encouraging more tie points to be matched), but it’s still really interesting to see what it produced.

Here’s an animated fly through of the densified point cloud and texture map (which is rather crude):

[youtube https://www.youtube.com/watch?v=dmFkOaMhbS4?feature=oembed&enablejsapi=1&origin=https://safe.txmblr.com&wmode=opaque&w=540&h=304]

I also ran Pix4D against the time slice film with Wuon Geon Ho:

[youtube https://www.youtube.com/watch?v=HkdPVbgmmPY?feature=oembed&enablejsapi=1&origin=https://safe.txmblr.com&wmode=opaque&w=540&h=304]

This computed 3D model was more similar to the video than the previous model was, but the entire back of the model didn’t make it. Still, the camera locations are plausible, and it’s really cool to be able to reconstruct the setup like this. If we had a measurement from the original scene (like the size of a book, or height of the model), we could measure actual working distances.

I imagine that the results would be much better with access to the original images vs using video frame grabs like I’ve done here.