HyperLapse Video Technology

[Werty]

Microsoft is introducing their new First Person HyperLapse Technology, pretty interesting.

Demo and tech videos, and a PDF.
http://research.microsoft.com/en-us/um/ ... yperlapse/

[snibgo]

It does look interesting. I have played with some of the concepts, for example stabilising a hand-held video from a person walking within a landscape, or a camera fixed to the front of a bike. Overall, this is a complex problem. In my simple scheme, I had good results by simply rotating images to match.

It seemed to me that the best solution was to construct a 3-D model using data from all the frames, and the video can then be reconstructed by "moving" the camera in any way we choose.

Which takes me back to some experiments around 1980, using hand-held photography to construct plans from scenes, by tilting the baseboard under the enlarger to correct the perspective.

[Werty]

I'd like to get one of those cameras that has built-in GPS and Compass, can be had for a reasonable price nowadays, even as Point&Shoot cameras, and then walk around and take hundreds of pictures of the same small area, from many different horizontal angles, and then import the pictures into a 3D engine like IrrLicht as textured planes, could be easily automated when the pictures include Location (GPS) and direction it's taken from (Compass) in the EXIF data.

All I would have to do is make sure all the pictures are taken horizontally, either on a tripod or a Pendulum screwed in the bottom of the camera.

Then move around in that 3D environment like in a first person shooter, all the pictures would be placed according to gps and direction taken, so when you walk towards and through one picture/textured plane, then the next would be visible.

There may be a problem with aspect ratio and perspective distortion, but could be cool to try anyways.

[snibgo]

I suspect that Google (or someone) publishes the software behind Google Street View (or similar).

[Werty]

That's no fun, I'd like to make it myself