Voting-based pose estimation for robotic assembly using a 3D sensor

Abstract

We propose a voting-based pose estimation algorithm applicable to 3D sensors, which are fast replacing their 2D counterparts in many robotics, computer vision, and gaming applications. It was recently shown that a pair of oriented 3D points, which are points on the object surface with normals, in a voting framework enables fast and robust pose estimation. Although oriented surface points are discriminative for objects with sufficient curvature changes, they are not compact and discriminative enough for many industrial and real-world objects that are mostly planar. As edges play the key role in 2D registration, depth discontinuities are crucial in 3D. In this paper, we investigate and develop a family of pose estimation algorithms that better exploit this boundary information. In addition to oriented surface points, we use two other primitives: boundary points with directions and boundary line segments. Our experiments show that these carefully chosen primitives encode more information compactly and thereby provide higher accuracy for a wide class of industrial parts and enable faster computation. We demonstrate a practical robotic bin-picking system using the proposed algorithm and a 3D sensor. IEEE International Conference on Robotics and Automation (ICRA)This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonprofit educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of Mitsubishi Electric Research Laboratories, Inc.; an acknowledgment of the authors and individual contributions to the work; and all applicable portions of the copyright notice. Copying, reproduction, or republishing for any other purpose shall require a license with payment of fee to Mitsubishi Electric Research Laboratories, Inc. All rights reserved. Abstract-We propose a voting-based pose estimation algorithm applicable to 3D sensors, which are fast replacing their 2D counterparts in many robotics, computer vision, and gaming applications. It was recently shown that a pair of oriented 3D points, which are points on the object surface with normals, in a voting framework enables fast and robust pose estimation. Although oriented surface points are discriminative for objects with sufficient curvature changes, they are not compact and discriminative enough for many industrial and real-world objects that are mostly planar. As edges play the key role in 2D registration, depth discontinuities are crucial in 3D. In this paper, we investigate and develop a family of pose estimation algorithms that better exploit this boundary information. In addition to oriented surface points, we use two other primitives: boundary points with directions and boundary line segments. Our experiments show that these carefully chosen primitives encode more information compactly and thereby provide higher accuracy for a wide class of industrial part...

Learning to Rank 3D Features

“…Choi et al [6] extended it using points on object boundaries and defined surface-to-boundary (S2B) and boundary-toboundary (B2B) pair features. They showed that for planar objects, pair features including boundary points encode more information for pose estimation and provide better performance.…”

“…Learning vs. uniform voting: We compared learning results with the baseline algorithm-uniform (equal) weights for each pair feature [7,6]. The results were computed for both synthetic and real data and are shown in Table 1 and Table 2, respectively.…”

“…2 Pose estimation using point pair features Figure 2 shows an overview of the voting-based pose estimation algorithm using point pair features [7,6]. A point pair feature consists of a pair of oriented points, denoted by their positions m r and m i and orientations n r and n i (e.g., normals or boundary directions).…”

“…Unless otherwise stated, in all the experiments using this dataset we used the B2B pair feature, which was shown to provide a good compromise between speed and accuracy [6], and we used weighting hash table bins scheme to learn the weights.…”

“…Overview of the voting-based pose estimation algorithm [7,6]. (a) A point pair feature is defined using two oriented points.…”