A Combined Texture-Shape Global 3D Feature Descriptor for Object Recognition and Grasping

This paper presents a global 3D feature descriptor for object recognition and grasping. The proposed descriptor stems from the clustered viewpoint feature histogram (CVFH) feature descriptor. Since the CVFH feature descriptor relies on shape information only, it obtains a poor performance when recognizing the objects with similar shapes. In order to improve the robustness and accuracy of object recognition, we extend CVFH feature descriptor with color information. Then this new global 3D feature descriptor is tested for multiple classes of 3D object classification by using support vector machines (SVM), and it is evaluated with a public dataset and real scenes respectively. The experimental results show that the proposed descriptor outperforms the CVFH feature descriptor in terms of recognition rate. Finally we utilize the proposed descriptor on our grasping system to recognize and grasp the objects, showing that the grasping system can accomplish the tasks well.