3D Hole Center and Surface Normal Estimation in Robot Vision Systems

In this paper, we present a novel stereo camera based 3D hole center and surface normal estimation method which can compute the relative pose (i.e., position and orientation) of a hole to a camera precisely. Unlike conventional methods which are completely dependent on either the geometrical model of a target or the visual feedback, the proposed method takes a novel open-loop approach on the basis of specific stereo image processing. Therefore, the computational speed of the proposed method is high enough to be applicable to real-time tasks in industrial applications while maintaining requirements for the localization accuracy. These breakthroughs are achieved by suppressing 3D geometrical errors from the stereo image processing based on the well-known DLT (Direct Linear Transformation) technique. In extensive experiments, the proposed method shows very promising performance in terms of both processing speed and pose estimation accuracy in a typical robot vision system which is equipped with a binocular eye-in-hand apparatus.