Abstract
Observers readily recognize actions in a dynamic (but not static) point-light display, suggesting that humans may identify biological motion using local motion signals. However, a point-light display provides configural information, albeit impoverished. I report two experiments using a novel stmulus with masks across apertures, demonstrating that when configural information is entirely eliminated, local motion information is insufficient to identify biological motion.
Participants were asked to identify walking direction of a human behind apertures. The stimulus consisted of 729 Gabor (drifting sine-wave gratings). Drifting velocity of Gabor was determined by a specified 2D motion of each element. Foreground elements lay on the trajectory of the walker; background elements off it. In Experiment 1, orientations of gratings were randomly assigned, eliminating configural information. When foreground Gabors drifted in the velocity consistent with 2D walker motion but background Gabors drifted randomly, participants were completely unable to identify walking direction (51% correct). Control conditions in which 2D motion of all background Gabors were the same (creating perception of a coherently moving background) yielded high accuracy (91%). Experiment 2 replicated the chance-performance condition of Experiment 1 but added configural information. Orientations of foreground Gabors were randomly assigned following a uniform distribution centered at the body structure orientation. Performance improved to 62%, 73%, 98% when the distribution range was ±80°, ±40°, ±10°, respectively. Biological motion is thus identifiable after adding even a small amount of noisy configural information.
Furthermore, randomizing 2D motion in foreground elements did not affect identification performance in both experiments. Similar results in Experiment 1 were also obtained using Plaids (two superimposed Gabors with orthogonal orientations), despite their reduced local motion ambiguity compared to Gabors. These findings suggest that local motion information serves to segment foreground against a noisy background, but additional configural information is necessary to identify biological motion.