paper which will address the user interface and 
visualization issues. 
Before continuing, a small theoretical digression is in 
order. One of the most useful concepts in defining 
quality in user interfaces is J.J. Gibson's theory of 
affordances (Gibson, 1966). This is theory of 
perception which states that we do not perceive 
patterns of light and shade, or lines and edges, or 
motion flow. Instead what we have evolved to 
perceive are 'affordances' which may be described as 
possibilities for action, or use. This we perceive 
surfaces as having the potential for walking or sitting, 
we perceive objects as potential tools or potential food 
and certain complex environments as holding potential 
danger. The theory embeds perception in action, and 
as such it makes sense of a good user interface as one 
in which the user perceives the right set of affordances. 
Through it, with minimal instruction the user can 
perceive the affordances of the computer system he or 
she is expected to use. The system should also afford 
the easy execution of the tasks for which it was 
designed. 
  
—@— 2D (tb) 
—i— Stereo perspective (1b) 
—&— Stereo head coupled perspective (1b) 
—&— Head coupled perspective (1b) ed X 
  
  
  
  
  
50 + 
  
  
  
  
  
% Error 
  
  
  
  
  
  
  
  
  
  
  
  
0 50 100 150 200 250 300 
Number of Nodes 
Figure 1. The results of a study of path tracing in an 
information network. Using a stereo, head coupled 
perspective view, as shown in figure 2. resulted in three times 
as many nodes being understood at the same error rate. 
We have recently obtained hard evidence that even for 
visualizing abstract information networks, where 
understanding the connectivity is important, visualizing 
in 3D is important (see Figure 1). However, it is not 
the fact that it is a perspective view that helps, but 
rather the enhanced space perception that comes from 
stereo viewing (which increases the size of the network 
that can be understood by 60%) and even more from 
motion parallax of the data (which increases the size by 
120%). If motion parallax is combined with stereo 
viewing we find that three times the network size can 
be understood for a constant error rate. 
488 
2. METHODS AND METAPHORS FOR 
VIEWPOINT NAVIGATION 
Viewpoint placement, 3D scene exploration and virtual 
camera control are all aspects of the same problem in 
computer graphics, namely how to move the viewpoint 
in a virtual 3D scene. The kinds of task where this is 
important are molecular modeling (Surles, 1992), 
walkthroughs of architectural simulations (Brooks, 
1986), camera control in animation systems, and 
flights over digital terrain maps representing subsea or 
remote sensing data (Stewart, 1991) as well as 
numerous CAD and advanced GIS applications. 
For a number of years we have been studying a six 
degree-of-freedom variant of the common mouse input 
device. We call it a Bat because a bat is like a mouse 
that flies (or fledermaus in German). The device 
senses both position (x,y,z) and orientation (azimuth, 
elevation and roll) information. In some studies we 
showed how this device could be used for object 
placement (Ware, 1990). However, more recently we 
have concentrated on using the Bat in ways that allow 
us to explore different methods and metaphors for 
virtual camera control. 
Often methods for viewpoint control are based on 
metaphors which help the user to get a conceptual 
grasp of the way the system will behave. Thus if the 
user is told that he or she is flying through the data, it is 
quite different than telling the user that the data is on a 
turntable which can be rotated. Most of the remainder 
of this paper is organized as a survey of different 
virtual camera control methods, both as employed in 
my research laboratory and by others. 
1) Eyeball in Hand Metaphor and Camera controllers. 
The phrase "Eyeball In hand" describes a Metaphor in 
which the user directly manipulates the viewpoint as if 
it were held in his or her hand. The metaphor requires 
that the user imagine a model of the virtual 
environment somewhere in the vicinity of the monitor. 
The eyeball (a spatial positioning device) is placed at 
the desired viewpoint and the scene from this 
viewpoint is displayed on the monitor. Cognitive 
affordance problems arise from the difficulty some 
subjects have of imagining the model. Ware and 
Osborne, (1990) found large individual differences in 
this respect. Also, if the eyeball is pointed away from 
the screen the correspondence between hand motion 
and the image motion is confusing. Physical 
affordances are restricted by the physical limitation of 
the device space - it can be awkward or impossible to 
place the "eyeball" in certain positions. 
There is a non direct-manipulation variation on this 
theme which allows for complex camera commands of 
the kind a director might give to the cameraman. 
Recent work by Gleicher and Witkin (1992) explores 
the use of high level commands to give the user control 
over the 
directing 
the direct 
certain mi 
or zoom | 
approach 
which pos 
the user tc 
al, 1992). 
2) World i 
In the "W 
changed t 
displayed 
rotated ck 
object on a 
Polhemus' 
rotations 
simultanec 
viewpoint 
single, rea 
the enviro 
enclosed ii 
does not 
affordance: 
center of rc 
interior the 
of being er 
hand moti 
(Ware and | 
The World 
the eyeball 
her viewpo 
Useful vari: 
virtual turni 
sphere (Ch 
from a mou 
direct man 
devices ten 
Often they 
the range o 
We are usi 
DEM visual 
3) Function 
It is a comn 
common gr 
translate anc 
directly by 
include this | 
it does not e 
A much mc 
functions is 
MacKinlay, 
which corre 
surface, hah 
also evaluat 
movements,