An Empiric Evaluation of Acknowledgment Methods for Optical See-Through Head Mounted Display Calibration

The calibration of optical see-through head-mounted displays (OSTHMDs) is an important fundament for correct object alignment in augmented reality. Any calibration process for OSTHMDs requires users to align 2D points in screen space with 3D points and to confirm each alignment. In this paper, we investigate how different confirmation methods affect calibration quality. By an empiric evaluation, we compared four confirmation methods: Keyboard, Hand-held, Voice, and Waiting. We let users calibrate with a video see-through head-mounted display. This way, we were able to record videos of the alignments in parallel. Later image processing provided base-line alignments for comparison against the user generated ones. Our results provide design constraints for future calibration procedures. The Waiting method, designed to reduce head motion during confirmation, showed a significantly higher accuracy than all other methods. Averaging alignments over a time frame improved the accuracy of all methods further more. We validated our results by numerically comparing the user generated projection matrices with calculated ground truth projection matrices. The findings were also observed by several calibration procedures performed with an OSTHMD.