First-person Hyperlapse video

From MS Research.


Would be interesting if this could actually Output a real 3D Scene?

That would be cool indeed. Although the complexity of the algorithm would be much higher.

Yeah … I infer they are mapping points that are in consecutive frames and see how they change from one to the next. They would then use that information do an automated Pan & Scan to reduce the jitter.

I don’t think they are building a full 3d model of the scene - but then again, what do I know :slight_smile:


I want to know how many hours they have invested in this…?

@ njbuch - Just in case. It is a different group at MS Research. Not the Gadgeteer people. ;D

I know. But still it could be nice to know.

I think you can’t build a real 3D scene. Your camera normaly only looks forward, so you only know one side off the scene. Only when rotation the camera all the time you get the missing informations. But also much more pictures to level out

If you want to build 3D scenes from pictures have a look at: