Fast and Robust Object Pose Estimation Based on Point Pair Feature for Bin Picking

Abstract

Random bin picking of industrial application is a complex and challenging task, where 3D object pose estimation based on point cloud is a key process. Recently, fast and robust object pose estimation algorithms have become an important concern for robotic bin picking. In this paper, an improved pose estimation pipeline for random bin picking is proposed based on point pair feature. In the improved pipeline, the point clouds are downsampled in an efficient way and a weight voting scheme is performed. A postprocessing for pose verification and multiple selection is also applied in bin picking application. Experiments on several synthetic datasets and real scenes demonstrate that the proposed method outperformed the original method and achieved competitive results in both recognition rate and time performance. The method in this paper can be applied to robotic random bin picking tasks with higher robustness and accuracy.