Pose Estimation of Sheet Stampings Based on Point Pair Features

Abstract

Robots are used in industrial production lines to automatically grab workpieces, and pose estimation is an important part in robotic bin-picking. However, existing pose estimation methods perform poorly on sheet stampings with weak texture, strong reflection, and multi-plane feature. Therefore, a pose estimation method based on point pair features with optimized preprocessing is proposed for the recognition of various plate-shaped metal parts. The proposed method performs planar segmentation and downsampling on the model point cloud in the offline stage, and selectively generates point pair features. A depth map filling algorithm is proposed to complete the depth information lost due to reflection. In order to reduce the interference of the placement plane, a distinguishing downsampling method is proposed to improve the efficiency. Experiments on multiple synthetic scenes and real scenes show that the proposed method has better performance in pose estimation of sheet stampings.