A 3D machine perception technique is presented using stereo intensity images and an anthropomorphic robot. Multiple stereo views allow to perceive the third dimension by the solution of a correspondence problem defined in two stereo pairs. The zero crossing technique is used to detect edges on the images and to classify them to reduce the number of possible solutions. By the eye-in-hand configuration several stereo pairs of scene can be taken in moving the robot arm. For a limited number of objects, still on the robot table, this technique allows the vision system to perceive a set of three-dimensional sample points that correspond to the physical object edges. The technique mainly consists in a geometrical match, in the 3D space, of some hypothesized solutions formulated on a couple of stereo pairs, taken from two points of view. Considering objects with straight edges and not highly textured surfaces, some geometrical properties are statistically verified for most of the solution points, so the correspondence problem can be robustly solved.
展开▼