Detection and Grasping of Texture-Less Objects Based on 3D Template Matching

Abstract

This paper studied a robust matching method using 3D point cloud for texture-less object detection and robotics grasping tasks. In view of the few surface feature points of texture-less objects, the template matching method based on Iterative Closest Point (ICP) improved LINEMOD algorithm is proposed to estimate the object pose. In our findings, the feature extraction of the LINEMOD algorithm is only carried out at the strong edges of the input image, and does not involve the calculation of internal texture features of the object, which effectively solves the problem of pose estimation for texture-less objects. However, only using the LINEMOD template matching is likely to cause mismatches, especially for objects with similar local contours. Thus, in order to make better use of the point cloud independent of object texture, a common point cloud registration optimization method with ICP is proposed. The ICP optimization happened between the model point cloud and the scene point cloud to improve the results of the pose detection with LINEMOD. Further, from the perspective of inverse kinematics of the manipulator, the object to be grasped has been positioned. The robotics experimental results show that our approach has good recognition and grasping accuracy for texture-less objects.