Evaluating stereo and motion cues for visualizing information nets in three dimensions

ACM Transactions on Graphics (1996), pp. 121-140, doi: 10.1145/234972.234975
Experiments
Tasks
Findings

The purpose of this experiment was to provide a quantitative estimate of the improvements in graph understanding offered by adding head coupling and stereo disparity cues to the display.

Variables and Constants

Independent Variables
Constants

Display components such as display form factor, jitter, resolution, among others, were held constant

System Info

Displays
  • 3d monitor - stereographics's crystaleyes 3d lcd stutter glasses
Input Hardware
Software
None

The computer is s Silicon Graphics Crimson VGX. I am assuming that this computer came with its default monitor, PowerVision Graphics, that supports stereoscopic viewing, hardware enabled texture mapping, 1280 x 1024 px, 24-bit color, (source: http://www.sgistuff.net/hardware/graphics/powervision.html) The Silicon Graphics Crimson VGX has specs approximately the following, in addition to the graphics specifications above: Processor: 100-150 MHz Cache: 8-16 KB, with a secondary cache of 1 MB Memory: 16-256 MB RAM Hard drive space: 1-2 GB Network: 4 Ethernet

Participant Info

Eleven subjects similar apparatus took part in this experiment, eight of whom had used and/or graphics systems before.

Total # Age Range Gender Balance
11 -

Two nodes in the graph were highlighted in red, and participants were asked whether they were linked by a path of 2 or none.

Interaction and Environment

Interface

The participant just has to select yes or no. Though the stereo is head-coupled, so the user could rotate their heads and look around.

Dimension varied in the study between a 2D graph and a 3D graph. Graph size, another variable, dictated the density of the environment.

Dimensionality Scale Density Visual Realism
3D Not reported Controlled variable Low
Metrics

  • errors - rate in which the participant was incorrect
  • time - time to answer the question
  1. There was a significant inverse effect of stereoscopy on errors for a comprehension of spatial information task.

    3D greaphs led to lower error rates in evaluating the path between two nodes than 2D graphs.

    Specificity: Neither
    The findings compared a stereo enabled, head-coupled system, with a 2D system. But the results were strongly significant, with p < .01

The purpose of this experiment was to provide a quantitative estimate of the improvements in graph understanding offered by adding head coupling and stereo disparity cues to the display. These are the 4 conditions: (1) 2D: No stereo, no rotation; the 3D graph was projected onto a 2D plane using an orthographic (parallel) projection by removing Z axis information, hence no overlap information was available. 2. Stereo perspective: No rotation; this condition made use of a pair of StereoGraphics CrystalEyes LCD shutter glasses to provide the disparity depth cues. 3. Head-coupled perspective The scene’s perspective projection changed continuously according to the subject’s measured head position; the perspective projection was defined by a single viewpoint centered between the eyes. 4. Stereo, head-coupled perspective: Same as the preceding, except with stereo. The correct view was generated for each eye position (continuously updated).

Variables and Constants

Independent Variables
Constants

Display components such as display form factor, jitter, resolution, among others, were held constant

System Info

Displays
  • 3d monitor - stereographics's crystaleyes 3d lcd stutter glasses
Input Hardware
Software
None

The computer is s Silicon Graphics Crimson VGX. I am assuming that this computer came with its default monitor, PowerVision Graphics, that supports stereoscopic viewing, hardware enabled texture mapping, 1280 x 1024 px, 24-bit color, (source: http://www.sgistuff.net/hardware/graphics/powervision.html) The Silicon Graphics Crimson VGX has specs approximately the following, in addition to the graphics specifications above: Processor: 100-150 MHz Cache: 8-16 KB, with a secondary cache of 1 MB Memory: 16-256 MB RAM Hard drive space: 1-2 GB Network: 4 Ethernet

Participant Info

This experiment involved 11 participants, 6 of whom had been exposed to a similar environment before. There was one person from Experiment 1 who took part in this one as well.

Total # Age Range Gender Balance
11 -

Two nodes in the graph were highlighted in red, and participants were asked whether they were linked by a path of 2 or none.

Interaction and Environment

Interface

The participant just has to select yes or no. Though the stereo is head-coupled, so the user could rotate their heads and look around. Dimension varied in the study between a 2D graph and a 3D graph. Graph size, another variable, dictated the density of the environment.

Dimensionality Scale Density Visual Realism
3D Not reported Controlled variable Low
Metrics

  • errors - rate in which the participant was incorrect
  • time - time to answer the question
  1. There was a significant interaction between stereoscopy and graph size on errors for a comprehension of spatial information task.

    Adding stereo decreases error rates as graph node size increases, compared to node cues, for tasks that involve graph understanding.

    Specificity: Neither
    Adding stereo allows for greater comprehension of graphs 1.6 times bigger, compared to seeing graphs with no visual depth cues. p < .01.

  2. There was a significant interaction between stereoscopy, graph size, and head tracking on errors for a comprehension of spatial information task.

    Adding head tracking gave decreased error rates as graph node size increases, more than adding stereo alone, for tasks that involve graph understanding.

    Specificity: Somewhat general
    Adding head tracking allows for greater comprehension of graphs 2.2 times bigger, compared to adding stereo alone. p < .01.

  3. There was a significant interaction between stereoscopy, graph size, and head tracking on errors for a comprehension of spatial information task.

    Adding head tracking & stereo gave decreased error rates as graph node size increases, more than adding head-tracking alone, for tasks that involve graph understanding.

    Specificity: Somewhat general
    Adding stereo & head tracking allows for greater comprehension of graphs 3.0 times bigger, compared to seeing graphs with no visual depth cues. p < .01.

Same question as before, with the following conditions: (1) 2D: No stereo, no rotation; the 3D graph was projected onto a 2D plane using an orthographic (parallel) projection by removing Z axis informa- tion, hence no overlap information was available. (2) Static perspective: No stereo, no rotation; essentially the same task as in (1), except that the graph is displayed using a perspective projection with the depth cues of relative size and overlap/occlusion. Conditions (2) through (9) all used a perspective projection. (3) Stereo: No rotation; this condition made use of a pair of Stereograph- ic CrystalEyes LCD shutter glasses to provide disparity depth cues. (4) Passive rotation: No stereo; the scene rotated at a constant angular velocity of 20 degrees/see about a vertical axis. (5) Stereo, passive rotation: Same as preceding except with stereo. (6) Hand coupled No stereo, hand coupled; lateral movement of the mouse caused rotation of the scene about a vertical axis; mouse movement towards and away from the subject caused rotation of the scene about a horizontal axis. Movement was restricted to 128° about the vertical axis and *49.1° tilt (about the horizontal axis). (7) Stereo, hand coupled: Same as preceding, except with stereo. (8) Head-coupled perspective: The scene’s projection changed continuously according to the subject’s head position; the perspective projection was defined by a single viewpoint centered between the eyes. (9) Stereo, head-coupled perspective: Same as preceding, except with stereo; the correct view was generated for each eye position. Fish Tank VR.

Variables and Constants

Independent Variables
  • head tracking - on or off, for fishtank vr mode
  • hand tracking - on or off, depending on condition
  • camera rotation - off, passive on with a constant velocity of 20 degrees/sec, or active on (with head or hand movements)
Constants

Display components such as display form factor, jitter, resolution, among others, were held constant. There were always 75 nodes present wit h100 arcs.

System Info

Displays
  • 3d monitor - stereographics's crystaleyes 3d lcd stutter glasses
Input Hardware
Software
None

The computer is s Silicon Graphics Crimson VGX. I am assuming that this computer came with its default monitor, PowerVision Graphics, that supports stereoscopic viewing, hardware enabled texture mapping, 1280 x 1024 px, 24-bit color, (source: http://www.sgistuff.net/hardware/graphics/powervision.html) The Silicon Graphics Crimson VGX has specs approximately the following, in addition to the graphics specifications above: Processor: 100-150 MHz Cache: 8-16 KB, with a secondary cache of 1 MB Memory: 16-256 MB RAM Hard drive space: 1-2 GB Network: 4 Ethernet

Participant Info

Once again, eleven subjects were used, Five of these subjects had had exposure to a similar set-up before, and there were two participants who were involved in both the last two experiments.

Total # Age Range Gender Balance
11 -

Two nodes in the graph were highlighted in red, and participants were asked whether they were linked by a path of 2 or none.

Interaction and Environment

Interface

The participant just has to select yes or no. Though the stereo is head-coupled, so the user could rotate their heads and look around. Dimension varied in the study between a 2D graph and a 3D graph. Graph size, another variable, dictated the density of the environment.

Graph size, another variable, dictated the density of the environment.

Dimensionality Scale Density Visual Realism
3D Not reported Controlled variable Low
Metrics

  • time - time to answer the question
  • errors - rate in which the participant was incorrect
  1. There was a significant interaction between stereoscopy, head tracking, hand tracking, and camera rotation on errors for a comprehension of spatial information task.

    Having no visual or motion cues led to higher error rates than in any combination of cues for graph comprehension tasks.

    Specificity: Somewhat general
    There was significance of P < .01 for 2D leading to higher error rates than all other conditions.

  2. There was a significant interaction between stereoscopy, head tracking, hand tracking, and camera rotation on errors for a comprehension of spatial information task.

    Adding motion or head/hand tracking cues to stereo led to lower error rates than stereo alone for graph comprehension tasks.

    Specificity: Somewhat general
    All pairwise comparisons between stereo and stereo + motion cues led to P < .01, except for stereo compared to stereo + head-tracking, which led to p < .05.

  3. There was a significant interaction between stereoscopy, hand tracking, and camera rotation on errors for a comprehension of spatial information task.

    Adding hand rotation to stereo led to lower error rates than adding passive rotation for graph comprehension tasks.

    Specificity: Neither
    Half of the pairwise comparisons with some sort of rotation or motion gave either p < .01 or p < .05. This finding is restricted to comparing active hand rotation + stereo and comparing it to all other motion conditions.

  4. There was a significant direct effect of camera rotation on time for a comprehension of spatial information task.

    Adding motion controls led to higher task completion times for graph comprehension graphs.

    Specificity: Neither
    Both the hand and head rotation/motion conditions were significantly slower than all the rest. There were no differences between them.

  5. There was a significant interaction between head tracking and camera rotation on time for a comprehension of spatial information task.

    Stereo, head-coupled condition led to slower task completion times than stereo, passive rotation condition for graph comprehension tasks.

    Specificity: Neither