Measuring and improving the accuracy of ARDev using a square grid

Augmented reality (AR) systems superimpose 3D virtual objects on top of a real scene. Most applications require the AR system to maintain an accurate registration between the virtual objects and the real scene. Many AR systems use parameters obtained from the camera calibration to set up a virtual camera that models the real camera. Thus, inaccuracy in camera calibration can adversely affect the registration of an AR system. This paper proposes a calibration improvement algorithm which measures and improves accuracy of a robotics AR system that is currently used in our lab. The algorithm uses a square grid pattern as a calibration device. The vertices of the grid are defined as calibration points. The image coordinates of the calibration points are extracted using image processing technique. By comparing the difference between these image coordinates and those that are calculated using the camera parameters, the accuracy of the system can be measured. The registration error found can be used to adjust the camera parameters accordingly and thereby improves the accuracy of the system. Experiments have shown that the calibration improvement algorithm is feasible and it can be used to measure and minimise the registration error of the AR system.