Robot Vision


If a robot arm is to be controlled by visual information from cameras, transformation between the image space and the world space has to be established. Conventional approach to solving this problem is based upon calibrating the vision system with the 3-D environment. the problem with calibration is tahat it is time consuming and it is useful only over a small subspaceof the robot's workspace. Another disadvantage is that the system must be recalibrated whenever the canera and/or the robot arm parameters are changed. In the last years, techniques that build 'weak' description of the 3-D space that does not require camera calibration, but is sufficient to perform visually controlled tasks, were developed.
In our case, the term 'weakly' calibrated system is used in a way that some constraints on the stereo camera positions against robot coordinate system are imposed, and that the orientation of the robot gripper during its movement is known. The two white stripes on the top edge of the gripper jaws with known lengths and mutual distance, serving as fiducial marks, and known pan and tilt angle, from the explicit knowledge that the control algorithem has about the system. Similar systems are in literature mostly addressed as uncalibrated systems, since no explicit calibration or self-calibration takes place.


Drago Torkar, Anton Pozne jr., Stanislav Kovačič, and Nikola Pavesic, "Weakly-calibrated Active Stereo Vision System", Proceedings of the Computer Vision Winter Wokshop CVWW'98, pp. 108-119, February 9.-12. 1998, Gozd Martuljek, Slovenia.