3D Object Recognition Using Line Structure

This paper presents an intelligent approach to recognize 3D objects using line structure correspondences. The proposed approach simultaneous recognizes an object and estimates the pose of the object. In order to achieve this goal, three challenges should be solved. First of all, line structures that human usually used to describe an object is used to represent the object. A set of such feature representation that shares the same properties with corresponding model line structures are first generated from images. Secondly, the structure correspondences are evaluated and ranked by additional features in the image. Only the most meaningful correspondences are selected. Each correspondence contributes a pose hypothesis with a transformation matrix. Finally, the approximate model pose hypotheses are estimated and refined based on the selected correspondences.