[Back to projects]

3D Interaction Research

I spent most of my senior year at UVa working on the undergraduate thesis A Comparison of Spatial Input to Alternative Techniques in 3-D Object Manipulation. This report claimed that current techniques for using (in this case rotating) graphical 3-D objects were inadequate for some applications and should be replaced with spatial input techniques. By spatial input techniques I mean devices, software, etc. that use freespace three-dimensional information from the user to effect changes on the graphical object. See [1].

To test this, I chose 4 different rotation techniques and implemented them wherever necessary. The techniques included:

  • Chen's Virtual Sphere [2] - a 2D "trackball" controller,
  • Shoemake's Arcball[3] - another trackball-style controller,
  • Hinckley's 3-D props[4], which are magnetically tracked 6-DOF objects, and
  • Three sliders for X, Y, and Z rotation.

I then wrote software to test users on a 3-D rotation task (see screen shot at right). The test is nearly identical to the one chosen in the Chen Virtual Sphere paper. The user is presented with two windows, one containing a still image and another with an interactive model of a 3-D house. The user's task is to match the interactive model with the orientation displayed in the still as accurately and quickly as possible for 10 images. I realize that time and accuracy can be a tradeoff, but interface-interface comparisons should not have been affected by individual users' thresholds for accuracy.


In any case, I tested about 25 users and took note of their anecdotal responses to the interfaces as well as the measurements made by my software, which were time to completion and accuracy (using the distance between quaternions). A majority of the users tested preferred using the props interface, with the Arcball and Virtual Sphere nearly even at second. The XYZ sliders interface served merely as a control test, forcing users to break up a 3-D rotation into its basic X, Y, and Z components, something which humans are known not to do well [5].

The figure below shows mean time to task completion for each interface. (Left to right, XYZ sliders, Arcball, Virtual Sphere, Props). Mean time ranges from 0 seconds to 45 seconds. This research became part of a later paper [6] and a chapter in Ken Hinckley's dissertation, and more detailed results can be found within them.

References:

  1. Hinckley, K., Pausch, R., Goble, J. C., Kassell, N. F., A Survey of Design Issues in Spatial Input, Proc. ACM UIST'94 Symposium on User Interface Software & Technology, April 1994, pp. 213-222. [PDF].
  2. Chen, M., Mountford, S.J., Sellen, A. A Study in Interactive 3-D Rotation Using 2-D Control Devices, Computer Graphics Vol.22 #4, 1988.
  3. Shoemake, K. ARCBALL: A User Interface for Specifying Three-Dimensional Orientation, Graphics Interface '92.
  4. Hinckley, K., Pausch, R, Goble, J., Kassell, N., Passive Real-World Interface Props for Neurosurgical Visualization, ACM CHI'94 Conference on Human Factors in Computing Systems, 1994, pp. 452-458. [PDF]
  5. Shepard, R., Metzler, J. Mental Rotation fo Three-Dimensional Objects, Science, 1971.
  6. Hinckley, K., Tullio, J., Pausch, R., Proffitt, D., Kassell, N., Usability Analysis of 3D Rotation Techniques, ACM UIST '97. [PDF]