Evidence for Good Recovery of Lengths of Real Objects Seen with Natural Stereo Viewing

Abstract

How good is human size constancy for real objects seen with natural stereo viewing, which minimises the opportunity for monocular size cues to play a role? This question has attracted renewed interest in recent years, arising mainly from the work of Todd and his colleagues. They have argued, initially from experiments in which stereograms were used, but more recently from studies based on real scenes, that poor performance on length judgment tasks suggests that human vision is weak at computing metric representations.

At ECVP '95, we described several experiments demonstrating quite good performance on the task of matching the lengths of two stationary real objects, gnarled wooden sticks, under binocular viewing with head held fixed (1995 Perception 24 Supplement, 129). We now report extensions to that work aimed at checking whether this good performance is maintained over three viewing distances (79, 158, and 355 cm), and when test and matching sticks are of different thicknesses. Matching performance was measured with a variety of indices: reliability, accuracy, Weber fraction, and absolute error. Relatively poor performance was observed when the sticks were viewed monocularly at the near and far distances but binocular viewing produced good performance at all distances. These results suggest that stereo can support good representations of metric scene structure when length judgments of natural objects are required under (quasi-)natural viewing. The implications of these results for theories of structure-from-stereo are discussed, and reasons are suggested why our results might differ from those of Todd and his colleagues.