Recognition of three-dimensional free-form objects using segment-based stereo vision

A method of recognition of free-form objects in a scene is described. By comparison of the established free-form object model and three-dimensional information about scene contours, the object is reconstructed according to segment-based stereo vision and its position and orientation are determined. At first, the scene contours are segmented to obtain its geometrical features, which are used in determination of an approximate candidate for its position. Such candidates are verified and adjusted against the contours of the object model to estimate their more precise positions and orientations. The free-form models are built from all-around range data of the object. Experimental results show the effectiveness of the method.