The Camera Calibration Procedure




In order to estimate the 3-D pose of the arm, we need to be able to predict what it will look like in a camera image. To do that, it is necessary to know

  • The location of the shoulder with respect to the camera
  • Some intrinisic parameters of the camera which define how it projects images onto the image plane.
    • An image of a special calibration pattern is captured:

    (calibration setup here)

    After clicking on the calibratoin pattern square corners, a least squares based camera calibration program is run. This program estimates the camera focal length, f, optical center, t, and radial distortion factor k. It also estimates the rigid transformation from the pattern's reference frame to that of the camera. With this information, it is possible to accurately compute where a point in 3-D will appear in the camera image.

    NOTE: We don't really need to know the transformation from the calibration pattern coordinate system to the camera's coordinate system. We could simply express 3-D coordinates in the camera reference system. But it helps to define the pattern coordinate system as the 'global coordinate system':

  • It is easier to define the shoulder location with respect to the pattern, rather than with respect to the camera coordinate system
  • If we move our hand-tip along the table top, it is easy to determine the ground truth for the hand-tip trajectory. We can then test the 3-D accuracy of the arm-tracking system.