Kinect: how do I convert depth image to pointcloud?

edit

The openni_kinect driver provides that functionality. It's broken up in nodelets - some are responsible for publishing the images, and some for processing them and assembling point clouds.

Normally, you start one nodelet manager, and all nodelets together on the same machine. In your use case, you would have to start two nodelet managers - one on for each machine on the end of the WiFi connection.

I had set up some launch files for a similar purpose a while back. You can find them here:

http://robotics.ccny.cuny.edu/data/kinect/launch_modular/

The openni_driver file is responsible for publishsing the image data, and the opneni_proc file is responsible for assembling it to 3D. They were set up for a single nodelet manager, so you would have to modify that.

The openni_record_player nodelet (which is in the semanticmodel package) does exactly this. It was developed to allow easy recording of Kinect data to disk, but should adapt to your use case pretty easily.