Abstract
A series of experiments is reported that examined the perception of the depth structure of a visual scene on the basis of binocular disparity and motion parallax information. Initial experiments (2.1-2.4) revealed that there are considerable differences in the perception of depth in computer simulated surfaces specified by each cue individually. These differences were interpreted as indicating a variation in the relative sensitivity of the visual system to different components of the geometric transformations generated between retinal images within the two domains. Subsequent experiments assessed observers' perception of depth, on the basis of disparity and/or parallax information, in configurations of point light sources in a dark limited cue environment viewed (i) directly (experiments 3.1, 3.2, 3.3 and 5) and (ii) through a head mounted CRT display (experiments 4.1-4.3). They showed that in these viewing conditions the visual system cannot derive a metric (Euclidean) representation of scene structure but it has access to a range of strategies (and consequently representations) that may enable it to complete tasks accurately. The strategy used depended upon the range of available information sources, an assessment of their reliability and the nature of the experimental task. Two final experiments examined the effect of increasing the available depth cues by (i) performing a task in an illuminated structured viewing environment and (ii) introducing surface texture cues. They showed that biases in depth perception persist in such environments and that they cannot be entirely explained by conflicting depth information signalled by accommodation (Frisby et al, 1996). It is argued that strong fusion models of depth cue interaction best describe the range of interactions found across of the all experiments. However, there are limitations on the types of strong interaction used by the visual system, i.e. no evidence was found for Richards' (1985) proposal that the simultaneous presence of disparity and motion information allow the recovery depth structure.