Pose Estimation Technique of Scattered Pistons Based on CAD Model and Global Feature

To estimate the pose of scattered pistons, an estimation technique was proposed based on the CAD model and global features. By automatic selecting the foreground point, the Min-Cut algorithm was improved from an interactive semi-automatic and binary-segmentation algorithm to a fully automatic and multiple-segmentation algorithm. The selection methods of viewpoint and cluster center of clustered viewpoint feature histogram (CVFH) were optimized. The point cloud of pistons was collected by Kinect, and the improved Min-Cut algorithm was used to segment the piston cloud. The optimized CVFH feature and the Camera Roll Histogram (CRH) feature of the pistons cloud were extracted, and the piston poses hypotheses were got by matching the features with the offline template library. The ICP algorithm was used to accurately match the pose. The false results were removed with the hypotheses verification and the six-degree-of-freedom posture of the piston was obtained. The average time of estimating the piston pose is 1.84s, and the average correct recognition rate is 98.61%. The proposed algorithm meets the requirements of grasping the pistons by the manipulator.