Abstract
Unfamiliar viewpoints can hinder visual object recognition from 2D static images. Here, we ask whether the same is true when visual input is in the form of dynamic spatio-temporal sequences, such as would accompany object or observer motion. Previous research has shown that such motion can be characteristic for a particular object and hence provide additional cues to identity. In two experiments we demonstrate that learned object motion can facilitate recognition across unfamiliar viewpoints. In each experiment, 24 participants were trained to discriminate between two novel amoeboid-like objects seen from a fixed viewpoint. These objects either deformed nonrigidly (Experiment 1) or rotated rigidly about a horizontal axis (Experiment 2). Both types of motion presented the observer with a coherent sequence of change that had a unique temporal order. After training, participants underwent a 2-interval-forced-choice task that tested their ability to discriminate the two learned objects from two novel objects. At test, objects were presented at 0°, 10°, 20° and 30° around the vertical axis relative to the learned viewpoint, and in the learned or reversed temporal order. The manipulation of temporal order has previously been used to study the contribution of motion to object recognition. In both experiments, accuracy decreased with increasing rotations away from the learned viewpoint and there was a constant benefit for learned object motion across all viewpoints tested (Experiment 1 = 4.9%; Experiment 2 = 5.3%). These results indicate that both rigid and non-rigid motion facilitated object recognition despite disturbances in 2D shape by viewpoint changes.