Abstract
While binocular disparity is well known for providing high-resolution discrimination thresholds, it also plays an important role in defining the separation of features or objects in depth. This suprathreshold performance has been assessed using a variety of techniques, most of which involve visual and/or haptic transformations. Ideally, if these techniques accurately assess depth percepts, they would be interchangeable, and equally affected by factors such as experience. To test this prediction we compared the accuracy of three depth estimation methods (haptic sensor, digital caliper, and a virtual ruler) using a simple line stimulus, with groups of experienced and naïve observers. Participants were asked to estimate the amount of depth between two vertical bars using each estimation technique. We found no consistent difference between measurements regardless of the method used. However, while experienced observers’ estimates followed geometric predictions, naïve observers consistently under-estimated small, and over-estimated large disparities. One explanation for this difference is that naïve observers are more sensitive to the cue conflict between stereopsis and perspective foreshortening. To test this hypothesis, a second group of naïve observers were assessed using the original and a perspective-corrected configuration. Our results showed a significant effect of removing cue conflict at the largest test disparity only. Closer examination showed that the data were bi-modal. Removal of the conflict eliminated the estimation errors for half of the observers; the remaining observers were unaffected by the manipulation. We conclude that the three techniques evaluated are equally accurate for the configuration used here. A more important consideration is the amount of experience with such procedures. While some observers readily discount conflicting depth cues, others do not and these individuals may require additional training. Failure to take experience into account will result in high inter-observer variability and distorted depth estimates.
Meeting abstract presented at VSS 2015