Building 3-D Visual Perception of a Mobile Robot Employing Extended Kalman Filter

The paper aims at designing a novel scheme for sensory data fusion by a mobile robot for reconstructing its 3-D world from their multiple gray images. Extended Kalman filter has been employed for determining the coordinates of the 3-D vertices and equation of the planes of the obstacles in the robot’s workspace from their multiple images. The geometric relations among these 3-D planes are then determined by using a logic program for recognizing the obstacles. The time required for recognition of a typical planer obstacle such as a box on a Pentium-III client with 64 MB RAM and a Pioneer-2 type robot server including the time involvement for the motion of the robot around the obstacle is approximately 18 seconds.