Abstract
Rapid Serial Visual Presentation (Potter, 1975) has been a central experimental paradigm for the determination of the temporal limits of visual object recognition. We present an extension of this technique for biological motion stimuli that requires subjects to detect specific human actions, which are embedded within a sequence of distractors. This new paradigm is used to investigate the temporal limits of the processing of biological motion based on form and spatio-temporal cues.
Method: Stimuli were based on randomly selected temporal segments drawn from point-light movies of 8 different actions (e.g., running, kicking, dancing, etc.). After presentation of the complete target action, subjects observed sequences consisting of 7 such temporal segments. Half of the sequences contained the target action, and the other half did not. Subjects had to indicate whether the target action was part of the presented sequence. The segments had durations of 6, 12 or 24 frames (corresponding to 200, 400 and 800 ms). In order to test the influence of the recognition of individual stimulus frames, sequences of static pictures with the same durations were also presented. These pictures were chosen randomly from the temporal segments.
Results and discussion: Recognition performance (d') with static frames was low and did not depend on the presentation time. This is consistent with Johansson's (1973) classical result that recognition of actions from static point light patterns is rather poor. Likewise, it seems difficult to match such patterns to a previously shown point-light stimulus. Performance for dynamic sequences is significantly better than for static stimuli already for 6 frames (200 ms). Performance for dynamic sequences increases strongly with the number of frames and levels off for 12 frames (400 ms). This suggests the existence of action-specific spatio-temporal pattern detectors that are optimally stimulated by short complex motion segments.
Supported by HFSP, Volkswagenstiftung and DFG.