Abstract
Immersive virtual reality is becoming an increasingly powerful tool for studying visuo-spatial perception in moving observers. However, the validity of results depends critically on an accurate calibration of the visual display. We have developed a system for calibrating a head mounted display (HMD) using camera calibration techniques (Gilson et. al., J. Neuroscience Methods, 173, 2008). Here, we describe a new method that allows a very large number of sample points to be gathered, leading to significant improvements in the accuracy of the recovered HMD frustum parameters. This improved accuracy allows us to test one of the key assumptions underlying the method. A camera was placed inside a stationary HMD where it recorded a chequerboard image shown in the HMD display. The position and orientation of the HMD was also recorded by a 6 degrees-of-freedom tracking system. Without moving the camera, the HMD was removed and images taken of a tracked calibration object that moved along a pseudo-random trajectory within the field of view of the camera. The technique allows camera coordinates of the calibration object to be translated to HMD coordinates. We used standard camera calibration techniques to recover the optical parameters of the HMD (not the camera) and hence derive appropriate software frustums for rendering virtual scenes in the binocular HMD. The recovered frustum has an optic centre coincident with the camera optic centre. We demonstrate the implications of this by introducing small shifts of the camera relative to the HMD. Our calibration method represents a significant advance over previous methods as it requires no detailed judgements by an expert user wearing the HMD (c.f. SPAAM, Tuceryan et al., Presence-Teleop. Virt., 11, 2002); it works for non-see-through HMDs; and it provides a quantitative measure of calibration accuracy, with reprojection errors of as little as 1 pixel (RMS).