3D object recognition using Kernel PCA based on depth information for twist-lock grasping

The handling of twist-locks has been a heavy burden for the container industry. There have been many efforts in developing automated twist-lock handling solutions. To address this challenge, we are developing a customized mobile manipulator for twist-lock pose estimation and grasping. In this paper, we propose a 3D object recognition approach using Kernel Principal Component Analysis (KPCA) only based on depth information to determine the basic information for twist-lock grasping using robotic manipulator. The challenge for twist-lock detection, recognition and grasping is 3D irregular object recognition in unstructured port environment. Motivated by gradient edge descriptor and KPCA, we propose a hybrid twist-lock detection approach without human intervention, in which we treat depth image as gray value image, and background difference method is combined with gradient edge descriptor. We also develop a set of kernel features on depth images, for description 3D object using kernel principal component features, to recognize types and pose of the twist-locks according to the nearest neighbor distance hierarchically. Experiments using a customized manipulator for detection, recognition and grasping twist-locks have been carried out to verify the feasibility of the proposed methods. Since depth images are insensitive to changes in lighting conditions, the proposed approach based on depth information is able to address the issues and solve problems caused by rust and painting peeled off of twist-lock handling in port environment.