Hand-eye Calibration using Images Restored by Deep Learning

Abstract

The problem of determining the transformation between the end-effector and the camera attached to the robot arm is called hand-eye calibration. For this calibration, the pose of the robot and the image data of the calibrated camera must be acquired. Since industrial robots use fixed focus cameras, the depth of field is limited, and variable focus lenses cannot determine the camera position or pose with high accuracy because the parameters of the camera change. If the pose of a robot is out of depth of field (DOF), the camera is out of focus, and the image is blurred. If hand-eye calibration is performed using these images, the resulting data includes errors. In this paper, image restoration is performed using deep learning on a blurred image where a marker cannot be found because it is out of focus. This technique is used to sharply restore images and demonstrate a technique for finding markers. We improved the precision of the hand-eye calibration problem by obtaining a clear image restored even in various pose of the robot and reducing the error that occurs in the blurred image.