Using ORK with manually calibrated stereo cameras.

edit

Hi,

Update: I have come across this question from the Google group for ORK and consequently asked this question hence wondered if simply mapping the left camera (of the stereo camera pair, since it produces pointcloud) to the depth camera topics of the ORK would do the trick ?

======

After reading the ORK tutorial, I understand that the key requirement for it is to have a depth registered (with RGB) image. This could be done using a manually calibrated stereo camera. However, the entire set of tutorials is based on a Microsoft Kinect ( which uses IR patterns ) and hence I am unsure about -

1. Is it possible to do the Object Capture phase using a standard stereo camera ? From that page I find a few missing details ( such as the size of the printed dot pattern not being required because the software internally computes it using a "Kinect-based plane finder" (please see this) to find the size ( I believe the size is needed for finding the 4 corners of the dot pattern in 3D coordinates (with z=0) and then using something like cv::solvePnP() to find the pose of the pattern relative to the camera.
2. Can all the phases (there are 3 - Data capture, training and detection) be done using a standard stereo camera pair, since the entire set of tutorials is based on a Kinect (which has a RGB and IR camera with fixed relative positions, and I noticed the Data capture phase stores the camera_info topic in a ROS bag (which may contain relative positions as horizontal stereo of the RGB and IR cameras of a Kinect and hence this doubt)
3. Lastly, if the 3 phases can be done using a stereo camera setup (think it's a rare possibility given the above), what kind of range can be expected ?
   • would that solely depend on the point cloud from the stereo cameras ?
4. Is the framework optimised for 0.3 - 5 metres (depth range of a Microsoft Kinect) ?

If someone can clarify any of the above, it would be very helpful.

Thanks!