The robot arm is an ongoing computer vision project for which many enhancements 
have been done during the last years, The aim of the paper is to present a system 
for controlling a robot arm, based on image processing and recognition. 
In the frame of this system, an important stage is the calibration. 
The presentation is focused on a new self-calibration algorithm based 
on video motion and color detection. The algorithm proves to be general enough 
to be employed in other similar computer vision scenarios where self-calibration 
might be useful. Calibration gets performed in a real-time having 
a low complexity per processed frame of O(n*log(n))