Non-deterministic method for semi-automatic calibration of smartphone-based OST HMDs

Abstract

An Augmented Reality (AR) system must show real and virtual elements as if they coexisted in the same environment. The tridimensional aligment (registration) is particularly challenging on specific hardware configurations such as Head Mounted Displays (HMDs) that use Optical See-Through (OST) technology. In general, the calibration of HMDs uses deterministic optimization methods. However, non-deterministic methods have been proposed in the literature with promising results in distinct research areas. In this work, we developed a non-deterministic optimization method for the semi-automatic calibration of smartphone-based OST HMDs. We tested simulated annealing, evolutionary strategy, and particle swarm algorithms. We also developed a system for calibration and evaluated it through an application that aligned a virtual object in an AR environment. We evaluated our method using the Mean Squared Error (MSE) at each calibration step, considering the difference between the ideal/observed positions of a set of reference points and those estimated from the values determined for the calibration parameters. Our results show an accurate OST HMD calibration for the peripersonal space, with similar MSEs for the three tested algorithms.