Thursday, June 2, 2011

Compression for 3D Video

This video shows an important step towards making 3D video conferencing work. The point cloud compression system uses Octomap and can detect spacial changes in real-time which allows only changes to be sent between key frames.

However, there will probably need to be additional work done to support compression of streaming 3D data from moving robots. Most likely the compression system will need to integrate with the localization system to achieve disambiguation between static points, changes in the known environment over time and new data available that becomes available as there are changes in visibility.


Robotbling said...

Very smart. It'll be super challenging to achieve the same thing with a moving robot, though, won't it?

I Heart Robotics said...

Getting it working on a moving robot will require some serious work, however it's the same work that needs to be done to achieve real-time 3D SLAM on-board the robot.

It will probably be implemented first on something like a TurtleBot or a BiliBot which can do 3D SLAM with four degrees of freedom, instead of six that are required for flying robots, by assuming the robot is in a planar environment and the roll and pitch of the robot are always zero the computational requirements can be decreased.

There are several groups working on the 3D SLAM problem and it is only a matter of development time and an iteration or two of Moore's law before it's ready for production.