This report is also available as an Acrobat file.

Contents
The Design of Virtual Environments with particular reference to VRML

The uses of spatiality

Movement perception (cont)

The optic array and optic flow

Many believe, with Gibson, that the major element in motion perception (and, perhaps, in visual perception generally) is the so-called optic flow (Figure 2).

Fig 2 On the left is the diagrammatic situation when we approach a surface at right angles to us. On the right, when we approach a surface at an angle, with the left edge nearest to us

As Harris (1994) perceptively comments:

As we move about and scan the visual world, the images of things move about, changing their relationships in a complex dance. JJ Gibson . . . was the first to understand this dance . . . Probably Gibson s greatest contribution was to redefine the dance-floor, emphasising the amount of information potentially available to an observer in the transforming optic array rather than the instantaneous fragments provided by a pair of retinal images.
The optic array is the three-dimensional bundle of light rays that impinge from all directions upon each point in an illuminated world. Objects in the world can be thought of as labelling specific rays, so producing a global pattern of light intensities. A retinal image provides access to only part of the optic array at any one time, but a stationary observer can sample different parts by eye movements and head rotations. By changing position, the observer can sample the different optic arrays impinging on neighbouring points in space. However, sampling in this case should not be thought of as a discrete process. Rather, as the observer gradually moves, so each ray gradually moves, thus producing the smooth transformation in the optic array that Gibson called the optic flow (p307-308).

Gibson believed that we picked up on the optic array and optic flow in order to understand the world. Unfortunately, he proposed no explanation as to how the picking up might be achieved. Johansson (1994) agrees with Gibson and suggests that the principles for decoding the optic flow are built-in to the specific visual systems of the species (p311). He outlines his approach to decoding the optic flow and what he calls the optic sphere in Johansson and BÜrjesson (1989).

Warren and Hannon (1988) conclude from studies of computer screen images that optic flow is sufficient at least for perceiving the direction of self-motion.

Many workers have made proposals for how we might cognitively deal with optic flow. Harris, Freeman and Williams (1992), already referred to, develop a concept by Koenderink and van Doorn (1976) who treat optic flow as a vector field in which the operations of translation, div (divergence), curl (rotation) and def (deformation, shear) apply (Figure 3).

Fig 3 showing the vector field operations of div, curl and def

De Bruyn and Orban (1990) give evidence to suggest that we are able to detect div, curl and def over much of the field of view.

Div, curl and def translate into operations on the optic flow lines as in Figure 4.

Fig 4 showing how div, curl and def affect the optic vector field

Dodwell (1983), basing his work on that of Hoffman (1996), uses the vector field ideas of Lie algebra to show how, with appropriate perceptual mechanisms, we might simply process optic flow (see also Hoffman 1970; Hoffman and Dodwell 1985). However, Caelli (1981 p143) takes the view that Hoffman s formulation is a meta-language that allows us conveniently to talk about some aspects of perception rather than giving us an explanation of it. He suggests that lower level models are needed before we can understand the situation. Harris, Freeman and Williams (1992) try to provide such models.

Biological motion detection

A curious ability we possess was first demonstrated by Johansson (1973a) who showed that, if small lights are placed on the ankles, knees, hips, wrist and elbows of a black-clad actor, and a movie is made of the person walking in a dark room where only the lights are visible, we can easily understand what we are seeing. Conversely, we cannot understand what we are seeing if no movement takes place.

This phenomenon has been studied by a number of workers. Kozlowski and Cutting (1977) show that the gender of an actor can be deduced from the moving pattern of lights. Runeson and Frykholm (1983) show that people are remarkably accurate in guessing the distance that the black-clad actors are able to throw small sandbags even though unable to see the bags or anything more than the lightspots on the actors joints. Pavlova (1992) showed that quite young children are able to recognise the patterns Berthenthal et al (1985) estimate that the ability to judge biological motion develops at around 6 to 9 months. Mather and West (1993) show that animals can also be recognised from this apparently limited information source. Sumi (1984) shows that, even if the film is run backwards and upside-down, subjects still recognise a human walker. What is more they perceive it as a walker with a peculiar gait rather than as an inverted image of a person moving backwards. Sumi concludes that this conception seemed to arise from the fact that the actor s arms were perceived as legs and vice versa. Dittrich (1993) shows that recognition of what is happening is more readily achieved when the figures are waking or running than carrying out social or instrumental actions. Further, that recognition is not greatly impaired if the lights are placed not on the joints but between them.

It is probably our ability to detect biological motion that explains the results of studies by Campbell (1979) who found that, in getting young children to understand still drawings of running and walking figures in implied motion, it was more important to show the figures in active postures than to employ motion cues like speed lines such as cartoonists sometimes use. Not surprisingly, Friedman and Stevenson (1975) found that older children were able to understand both postural cues and the cartoonists conventions. They too, however, seemed to favour active postures.

Movement in graphics and virtual reality

There is little doubt that movement helps in our understanding of objects. Frequently, drawings that make little sense when shown as stills are understandable when they are animated. This is particularly the case in viewing wire-frame drawings. Wire frame drawings of three-dimensional scenes often contain too many lines for us to disambiguate when seen in still form. When animated to allow for panning across the scene and rotating objects, however, they spring to life and the third dimension appears effortlessly. Often, indeed, a wire-frame animation seems to convey more information about the form of an object than does a movie of the same thing (although, of course, it cannot tell us anything about the surface properties of the object).

The attractions of three-dimensional drawings, particularly animated three-dimensional drawings, are such that, sometimes, they are used inappropriately for example, to illustrate two-dimensional data. When this happens not only do they not enhance the data set, they sometimes falsify it. We should also note that, although we seem to have separate mechanisms for dealing with motion, colour, and form (at least), it is clear that we can better understand subtle changes in a scene when we attend to one of these aspects only (Corbetta et al 1990). From this, and other studies, we would conclude that, if we wish to have someone focus their attention on a particular element in an image or animated scene, then we should not create conflict by changing motion, colour and form together. This would be especially the case if abstract data is being shown.

Interestingly, in an admittedly slightly limited and uncontrolled study by Felix (1995), we find that there is no evidence that being, as it were, inside a virtual space and walking about it, as opposed to viewing it from outside from different viewpoints, necessarily improves understanding of it. In both cases, motion assists. But immersion in the space does not necessarily provide good understanding of it.

Summary

The detection of motion is a strong element in visual perception and we gain much information from it. Indeed, some believe motion to be the prime source of our understanding of the world. Sometimes, it is only through motion that we can disambiguate conflicting signals and it is likely that our ability to process optic flow is primal.

Contents

Graphics Multimedia Virtual Environments Visualisation Contents