Verification of recognition performance of cloth handling robot with photo-model-based matching

Abstract

Recently, vision-based robots have been used in the garment factories. However, deformable objects such as cloths, strings have been difficult to be recognized and handled automatically by robots. Moreover, robots have been facing with two main problems in 3D cloth recognition and handling performance. These problems are to judge that the cloth placed in front of the robot is rightly the intended one to be handled and to pick and place (handle) at a designated position automatically. In this paper, model generation method from cloth photograph and model-based matching method (recognition method) utilizing Genetic Algorithm (GA) are presented. The proposed system recognizes the target cloth and estimates the pose of that cloth for handling even though the target cloth is occluded partially and there are another cloth in the field of view of robot's camera. In this paper, 3D recognition accuracy has been confirmed experimentally by using 12 different samples cloths.