Magnus Axholt

Label segregation by remapping stereoscopic depth in far-field augmented reality

This paper describes a novel technique for segregating overlapping labels in stereoscopic see-through displays. The present study investigates the labeling of far-field objects, with distances ranging 100-120 m. At these distances the stereoscopic disparity difference between objects is below 1 arcmin, so labels rendered at the same distance as their corresponding objects appear as if on a flat layer in the display. This flattening is due to limitations of both display and human visual resolution. By remapping labels to pre-determined depth layers on the optical path between the observer and the labeled object, an interlayer disparity ranging from 5 to 20 arcmin can be achieved for 5 overlapping labels. The present study evaluates the impact of such depth separation of superimposed layers, and found that a 5 arcmin interlayer disparity yields a significantly lower response time, over 20% on average, in a visual search task compared to correctly registering labels and objects in depth. Notably the performance does not improve when doubling the interlayer disparity to 10 arcmin and, surprisingly, the performance degrades significantly when again doubling the interlayer disparity to 20 arcmin, approximating the performance in situations with no interlayer disparity. These results confirm that our technique can be used to segregate overlapping labels in the far visual field, without the cost associated with traditional label placement algorithms.

Parameter estimation variance of the single point active alignment method in optical see-through head mounted display calibration

The parameter estimation variance of the Single Point Active Alignment Method (SPAAM) is studied through an experiment where 11 subjects are instructed to create alignments using an Optical See-Through Head Mounted Display (OSTHMD) such that three separate correspondence point distributions are acquired. Modeling the OSTHMD and the subjects dominant eye as a pinhole camera, findings show that a correspondence point distribution well distributed along the users line of sight yields less variant parameter estimates. The estimated eye point location is studied in particular detail. The findings of the experiment are complemented with simulated data which show that image plane orientation is sensitive to the number of correspondence points. The simulated data also illustrates some interesting properties on the numerical stability of the calibration problem as a function of alignment noise, number of correspondence points, and correspondence point distribution.

Visual clutter management in augmented reality: Effects of three label separation methods on spatial judgments

This paper reports an experiment comparing three label separation methods for reducing visual clutter in Augmented Reality (AR) displays. We contrasted two common methods of avoiding visual overlap by moving labels in the 2D view plane with a third that distributes overlapping labels in stereoscopic depth. The experiment measured user identification performance during spatial judgment tasks in static scenes. The threemethods were compared with a control condition in which no label separation method was employed. The results showed significant performance improvements, generally 15–30%, for all three methods over the control; however, these methods were statistically indistinguishable from each other. Indepth analysis showed significant performance degradation when the 2D view plane methods produced potentially confusing spatial correlations between labels and the markers they designate. Stereoscopically separated labels were subjectively judged harder to read than view-plane separated labels. Since measured performance was affected both by label legibility and spatial correlation of labels and their designated objects, it is likely that the improved spatial correlation of stereoscopically separated labels and their designated objects has compensated for poorer stereoscopic text legibility. Future testing with dynamic scenes is expected to more clearly distinguish the three label separation techniques.

Technical Section: Objective and subjective assessment of stereoscopically separated labels in augmented reality

We present a new technique for managing visual clutter caused by overlapping labels in complex information displays. This technique, label layering, utilizes stereoscopic disparity as a means to segregate labels in depth for increased legibility and clarity. By distributing overlapping labels in depth, we have found that selection time during a visual search task in situations with high levels of visual overlap is reduced by 4s or 24%. Our data show that the stereoscopically based depth order of the labels must be correlated with the distance order of their corresponding objects, for practical benefits. An algorithm using our label layering technique accordingly could be an alternative to traditional label placement algorithms that avoid label overlap at the cost of distracting view plane motion, symbology dimming or label size reduction.

Optical See-Through Head Mounted Display Direct Linear Transformation Calibration Robustness in the Presence of User Alignment Noise

The correct spatial registration between virtual and real objects in optical see-through augmented reality implies accurate estimates of the users eyepoint relative to the location and orientation of the display surface. A common approach is to estimate the display parameters through a calibration procedure involving a subjective alignment exercise. Human postural sway and targeting precision contribute to imprecise alignments, which in turn adversely affect the display parameter estimation resulting in registration errors between virtual and real objects. The technique commonly used has its origin in computer vision, and calibrates stationary cameras using hundreds of correspondence points collected instantaneously in one video frame where precision is limited only by pixel quantization and image blur. Subsequently the input noise level is several order of magnitudes greater when a human operator manually collects correspondence points one by one. This paper investigates the effect of human alignment noise on view parameter estimation in an optical see-through head mounted display to determine how well a standard camera calibration method performs at greater noise levels than documented in computer vision literature. Through Monte-Carlo simulations we show that it is particularly difficult to estimate the users eyepoint in depth, but that a greater distribution of correspondence points in depth help mitigate the effects of human alignment noise.

Evaluation of Alternative Label Placement Techniques in Dynamic Virtual Environments

This paper reports on an experiment comparing label placement techniques in a dynamic virtual environment rendered on a stereoscopic display. The labeled objects are in motion, and thus labels need to continuously maintain separation for legibility. The results from our user study show that traditional label placement algorithms, which always strive for full label separation in the 2D view plane, produce motion that disturbs the user in a visual search task. Alternative algorithms maintaining separation in only one spatial dimension are rated less disturbing, even though several modifications are made to traditional algorithms for reducing the amount and salience of label motion. Maintaining depth separation of labels through stereoscopic disparity adjustments is judged the least disturbing, while such separation yields similar user performance to traditional algorithms. These results are important in the design of future 3D user interfaces, where disturbing or distracting motion due to object labeling should be avoided.

User boresight calibration precision for large-format head-up displays

The postural sway in 24 subjects performing a boresight calibration task on a large format head-up display is studied to estimate the impact of human limits on boresight calibration precision and ultimately on static registration errors. The dependent variables, accumulated sway path and omni-directional standard deviation, are analyzed for the calibration exercise and compared against control cases where subjects are quietly standing with eyes open and eyes closed. Findings show that postural stability significantly deteriorates during boresight calibration compared to when the subject is not occupied with a visual task. Analysis over time shows that the calibration error can be reduced by 39% if calibration measurements are recorded in a three second interval at approximately 15 seconds into the calibration session as opposed to an initial reading. Furthermore parameter optimization on experiment data suggests a Weibull distribution as a possible error description and estimation for omni-directional calibration precision. This paper extends previously published preliminary analyses and the conclusions are verified with experiment data that has been corrected for subject inverted pendulum compensatory head rotation by providing a better estimate of the position of the eye. With correction the statistical findings are reinforced.

Baseline SPAAM calibration accuracy and precision in the absence of human postural sway error

We conducted an experiment in an attempt to generate baseline accuracy and precision values for optical see-through (OST) head mounted display (HMD) calibration without the inclusion of human postural sway error. This preliminary work will act as a control condition for future studies into postural error reduction. An experimental apparatus was constructed to allow performance of a SPAAM calibration using 25 alignments taken using one of three distance distribution patterns: static, sequential, and magic square. The accuracy of the calibrations were determined by calculating the extrinsic X, Y, Z translation values from the resulting projection matrix. The standard deviation for each translation component was also calculated. The results show that the magic square distribution resulted in the most accurate parameter estimation and also resulted in the smallest standard deviation for each extrinsic translation component.

Comparing disparity based label segregation in augmented and virtual reality

Recent work has shown that overlapping labels in far-field AR environments can be successfully segregated by remapping them to predefined stereoscopic depth layers. User performance was found to be optimal when setting the interlayer disparity to 5--10 arcmin. The current paper investigates to what extent this label segregation technique, label layering, is affected by important perceptual defects in AR such as registration errors and mismatches in accommodation, visual resolution and contrast. A virtual environment matched to a corresponding AR condition but lacking these problems showed a reduction in average response time by 10%. However, the performance pattern for different label layering parameters was not significantly different in the AR and VR environments, showing robustness of this label segregation technique against such perceptual issues.