The Icosahedron Marker for Robots 6-Dof Pose Estimation

Abstract

ArUco is an efficient and convenient method to estimate pose of the robot or the object by detecting corner points. However, ArUco cannot be used in some cases due to its lack of detection stability and insufficient accuracy. In this paper, an improved monocular 6D pose estimation method based on the spatial icosahedron marked by ArUco is proposed. By identifying the marked points on the surface of the icosahedron in the image, matching the 3D points in the database and solving with the Perspective-n-points(PnP) method, the accurate 6D pose estimation of the robot can be obtained. Compared with the original ArUco, the method proposed in this paper can cope with larger pose changes, provide higher accuracy, and improve the detection stability, while the time complexity is not significantly increased. In most cases, the method proposed in this paper can achieve translation accuracy of sub-millimeter level and rotation accuracy of less than 1°. Besides, it realized the ±180° large-scale rotation detection of three axes, and the real-time indicators is also satisfied.