MODEL BASED 3D OBJECT RECOGNITION USING LINE FEATURES1

Abstract – This paper presents an effective and robust model-based 3D object recognition algorithm using line feature correspondences. The algorithm identifies an object and estimates the pose of the object and provides the object recognition information to a home service robot system. The identification of the object and estimation of the poses is done by a three-stage processing: In the first stage, a list of corresponding image line sets that share same properties of mode line sets is generated. In the second stage, the correspondences in the list are evaluated and ranked. Only a small number of correspondences from the list are selected. Each correspondence contributes a pose hypothesis consisting of a similarity transformation. In the last stage, the approximate model pose hypotheses are estimated based on the selected correspondences. Local adjustment is used to ensure the matching and to improve pose accuracies