The perception of depth, rotation and shearing in motion parallax surfaces
Motion parallax is often considered to be an inherently ambiguous cue to depth. Despite the theoretical ambiguity associated with the pattern of retinal image motion, motion parallax generally evokes compelling three-dimensional (3-D) percepts and for this reason is regarded as an important source of 3-D information. Certain studies have indicated, however, that a parallax surface that contains a given amount of simulated depth is often perceived to rotate, rather than simply remain stationary, as the observer moves. This thesis provides an experimental investigation of the factors which influence perceived rotation and shearing in motion parallax surfaces. In a series of psychophysical experiments, the cue of self-produced motion parallax was manipulated in order to provide insights into the mechanisms underlying the perception of 3-D surfaces. Since larger parallax motions often produce the impression of rotation, the "transition point" between stationarity and rotation was measured as a function of several factors. The maximum motion gradient was shown to be the principal determinant of this transition point surfaces with a steep motion gradient were perceived to rotate at lower relative motion amplitudes than surfaces with shallow motion gradients. Vertical perspective information played a smaller role. The transition point also fell with increasing viewing distance. At even higher amplitudes, parallax surfaces can appear nonrigid or even lacking in all 3-D structure, and the experiments reported have measured the transition points between each of the different perceptual zones. A model was introduced in order to determine whether the perceived magnitudes of depth and rotation of sinusoidal parallax surfaces were in accordance with geometric constraints. Qualitative support was found for a trade-off between depth and rotation when corrugation frequency or stimulus size was manipulated. Results were less conclusive when a dynamic vertical perspective cue was varied. A similar model was applied to the perceived magnitudes of depth and shearing in square wave surfaces. The relationship between these attributes was less clear-cut. The perceived rotation of sinusoidal surfaces increased with increasing viewing distance; possibly due to a decreasing propensity of the observer, with increasing viewing distance, to attribute the vertical perspective changes to self-motion. In sum, these experiments demonstrate the importance of motion gradients and vertical perspective information in the perception of motion parallax surfaces, and suggest that the surfaces are generally perceived in qualitative accord with the totality of visual information present (Helmholtz, 1909).