Colloquium: What is Stereo Vision good for?

The natural world provides a rich and varied array of depth information – numerous cues that are commonly divided into monocular and binocular categories. Our experience with two-dimensional representations of depth in paintings, 2D movies and computer simulations is so vivid that the additional information supplied by the primary binocular cue, namely stereopsis, might seem superfluous. What does stereo add to our knowledge about the location and shape of real objects? To answer this question, we asked subjects to judge the depth between two real objects, viewed either monocularly or binocularly, and measured their sensitivity to small incremental changes in depth. Even in an enriched environment containing texture, occlusion and motion parallax cues, the monocular thresholds were very imprecise, ranging from 5 to 18 times worse than the binocular thresholds. Consistent with these results, recent work from other laboratories (Melmoth et al, 2009) shows that grasping movements are slower and less precise during monocular viewing. Thus, the main contribution of stereopsis in natural settings is a highly precise, if biased, representation of the distance separating objects along the line of sight. Surprisingly, some subjects with excellent stereoacuity for natural objects were unable to make precise judgments of pure disparity in stereoscopic computer displays, which suggests that, in naïve subjects, conflict from imprecise monocular cues can degrade even the most reliable stereo information.