Stimulus-dependent depth constancy during head tilt

Stereopsis is traditionally measured with noise-based stereo tests while the observer views the test in primary gaze. We investigated the effect of stimulus sparseness and axial variations of interocular disparity induced via head rotations. First, we measured stereoacuity using a 4-Alternative-Forced-Choice (4-AFC) task with three uncrossed and one crossed disparity bandpass-filtered circles on a passive-3-D-monitor. Ten binocularly-normal adults fixated a central cross and clicked on the circle withcrossed disparity for forty trials/condition. Observers adopted head tilts of 0° or ±20° pitch, roll, or yaw, enforced with an innertial measurement unit and fixation enforced with an eye tracker. Next, we measured stereoacuity in 8 adults while either the head (H), monitor (M), or both (B) were tilted 0°, ±22.5°, or ±45° roll in random order (eighty trials/condition) using a 4-AFC task and random-dot stimuli. Head tilts did not signifcantly alter stereoacuity using narrow-band stimuli(p>0.05), despite that IPDs and the axis of disparity were differentially affected by the tilts. However, for random dot stimuli, stereoacuity decreased with increasing orientation difference between the head and monitor (H and M: p<0.05; B: p>0.05]. Head tilt decreases IPD and rotates the axis of interocular disparity, however, these manipulations affect stereoacuity when measured with noise stimuli but not with sparse stimuli. The results are consistent with a vestibular input to stereoscopic disparity processing that can be detected by sparse stimuli but is masked by dense stimuli. The results have implications for natural vision and for clinical screening in patients with abnormal head posture. Significance statement Depth perception is a critical feature of human vision and it is thought that the ability to perceive stereoscopic depth is bound to an essentially eye-fixed, horizontal disparity of each image that rapidly deteriorates away from that limited horizontal axis. In a set of head tilt experiments, we varied the orientations of stereoscopic images and demonstrate that stereoacuity remains constant when deploying sparse narrow-band stimuli, and only worsens when using fine-detailed noise stimuli that mask off-axis disparities. These results shine new light upon the debate of neuroplasticity of stereo vision. Moreover, the results are consequential for diagnosis and treatment in people with atypical head- and eye alignment, such as for patients with torticollis or strabismus.

customized C++ program ran simultaneously with MATLAB 2019a using Psychtoolbox-3 and Open-GL to 136 calculate the head location and orientation in real time. Feedback concerning 3D head orientation was 137 presented in real-time to the observer on the experimental display via a 3D 6° colored cube at the center of 138 the screen controlled with Open GL software and by a 2° white head-posture-contingent ring that moved 139 with the observer's head. The target head posture was indicated with a wire frame cube and by a 6° green 140 ring. To obtain the required head posture, the observer's task was to move their head to align the cube with 141 the wire frame (at which point the cube changed from red to green), and simultaneously moved the white 142 ring so that it was within the green circle (see Figure 3). Observers were required to hold the required head 145 Eye movement control 146 Binocular eye movements were measured simultaneously with Tobii EyeX (Sweden) remote eye 147 tracker with a sampling rate of 60 Hz, using the Tobii EyeX Matlab toolkit (12). The observer's 148 gaze task was always to look directly at a fixation point at the center of the screen, regardless of the 149 head posture required for any condition. Real-time feedback concerning gaze direction for each 150 eye was provided with two small circles, one for each eye. The observer's task was to maintain the 151 gaze-contingent rings within 3° of the fixation target (see Figure 3). Once observers held this gaze 152 and the required head posture for at least 800msec, an experimental trial was initiated.

Monitor and computer
154 Stimuli were presented on a polarized 42" 3D TV from LG (model 42LM6200-UE) with refresh 155 rate of 60 Hz at a resolution of 1920 by 1080 pixels. Dichoptic viewing was obtained with LG 156 cinema 3D glasses (AG-F310), over spectacles if worn. Crosstalk was minimized with 157 Psychtoolbox's StereoCrosstalkReduction and SubtractOther routines to minimize the subjective 158 visibility of a 100% contrast 4 c/deg sine grating presented to one eye and a mean luminance field 159 presented to the other eye. The monitor was gamma-corrected prior to the experiment using a 160 Photo Research SpectraScan 655 (Norway) spectrophotometer. The viewing distance was set to 161 120 cm from left eye's corneal apex to the screen during the upright baseline condition. 182 We therefore implemented an initial screening step (15), to individualize the starting parameters of 183 the staircase for each observer. Six bandpass-filtered circles were horizontally positioned with a 184 center-to-center distance of 2.5° ( Figure 2A). The diameter of each circle was 1.25° and was 185 filtered with a log cosine filter with a peak spatial frequency of 4cyl/° and Full Width at Half- 255 where p Guess was 25% (4AFC task), x is disparity, and threshold and slope were free fit parameters.
256 Stereothresholds, here refered to as stereoacuities for 7 headtilt conditions were calculated for each 257 of the ten participants and were stored in an 7*10 matlab array. One-way Analysis of Variance 258 (ANOVA) using Matlab 2021a was deployed to compared results for an effect of headtilt condition.  329 The mean stereoacuity, across subjects and conditions was 122 arcsec ±3.5 standard deviation and 330 did not significantly vary with head posture (see Figure 7). This estimate is in line with typical 541 AC and PJB wrote the code for the psychophysical procedure of first experiment.
542 AC recruited participants and collected data for the first experiment.
543 JS, AC, PJB analysed the data for the first experiment.
544 JS and PJB designed the protocol of the second experiment.
545 JS recruited participants and collected the data for the second experiment.
546 JS and PJB analysed the data for the second experiment.