Abstract
Transparent motion stimuli allow us to investigate how visual motion is processed in the presence of multiple sources of information. We used stereo random-dot kinematograms to determine how motion processing is affected by the difference in direction and depth of two overlapping motion components. Observers judged whether a noise dot display contained one or two directions of motion. For all disparity differences, performance did not change among angles greater than 60 degrees, but the ability to detect transparent motion fell dramatically as the direction difference decreased below 60 degrees. When a disparity difference was added between the two motion components, detection became easier. We compared these results to an ideal-observer model limited by stimulus uncertainty and low-level sources of internal noise. The resulting measure of efficiency--the ratio of human to model performance--reflects changes in how motion stimuli are being processed. A decrease of both the direction and disparity differences had the effect of decreasing efficiency. These results suggest that the mechanism processing transparent motion may implement a smoothness constraint that tends to combine similar motions into a single percept.