An efficient two-step solution for vision-based pose determination of a parallel manipulator

A vision sensor for the direct pose determination of a parallel manipulator can extend the manipulators capabilities in many aspects. The existing approaches to a solution to the pose problem fall into two distinct categories: analytical solutions and iterative solutions. In this paper, we present an efficient two-step solution for the pose determination of a parallel manipulator by combining the advantages of the above two methods. Four points in an image, vertices of a parallelogram in object space, and the correspondences among these four coplanar points are utilized in the proposed method. In the first step, four coplanar vertices of a parallelogram, which locate the parallel manipulator's surface, are detected, Subsequently, the pose of the parallel manipulator can be determined with two affine invariants. In the second step, an iterative method is introduced, and the results of the first step are passed to the current step being used as initial values of the iterative process. Also, an error matrix is established through seven error functions, that are produced by the depth estimation and the co-planarity of the four vertices. The proposed method has been experimentally validated by a calibration board and the Pan and Tilt platform, which is a parallel manipulator developed by Googol Technology Ltd. The proposed method is then compared with three other existing methods. Experimental results demonstrate that the proposed method has the advantage of lower computational cost. The accuracy and stability, which are the main concerns during the real-time applications, are also improved.

► An efficient two-step solution for the pose determination of a parallel manipulator.
► The solution combines the advantages of analytical solution and iterative solution.
► The pose parameters are first estimated by a simple analysis solution.
► The analytical results are used as the initial values of the iterative process.
► The method processes each image frame separately with no calculation error.