Real-time humanoid visual system based on region segmentation

This paper presents a visual system for a humanoid robot. The eye mechanism possesses five degrees of freedom; its dimensions and motion behavior imitate real human eye. The software combines the techniques of image segmentation, region extraction, region correspondence, and projective geometry of camera to achieve the function of distance determination. Unlike general fixed stereo system, the errors from eyeball rotation are also considered. An angle control algorithm is implemented in the system to reduce the errors. This system could be applied in obstacles detection and objects capture.