Robust Object Pose Estimation Based on Improved Point Pair Features Method

Abstract

In the robot grasping system, it is necessary to estimate the pose of the object. The common method of 3D pose estimation is to use the CAD model of the original object. However, sometimes the 3D model of the object is not easy to obtain. We propose a new method, which obtains the three-dimensional point cloud of the template object captured by depth camera. The improved method filters out redundant candidate poses, and uses the proposed pose verification method to improve the accuracy of pose estimation. Compared with the original PPF method, we get a better recognition rate in a large number of workpieces pose recognition tasks.