Abstract
Searching for a person in a crowd is an important task. But arguably, an even more important task is finding (and evading) a fighting person in a crowd. Biological motion research, however, has been heavily tilted towards the use of walking figures. Here we utilized a visual search paradigm to examine what information is essential in detecting abnormal activities.
We employed two target-distractor combinations (a boxer target among walkers distractors, and vice versa), and three different set sizes (3, 6, or 9 items). Observers indicated the presence or absence of a target action. The first block included intact point-light actors; the second block was composed of scrambled actors in which each joint's initial position was spatially scrambled. All point-light actors rotated in depth.
We observed a search asymmetry between boxers and walkers: Searching for a boxer among walkers was faster and more efficient than searching for a walker among boxers. In fact, the boxer popped out, suggesting the search for boxing actions occurred preattentively. Configural information cannot explain the search asymmetry, because the same effect was obtained with scrambled point-light actors. ROC analyses showed that local motion signals (i.e., average and maximum velocity and acceleration) could not explain this effect, as they neither showed the search asymmetry, nor reached human performance level. However, reverse correlation techniques showed that the observers' responses did cluster around the “punch” event of a boxer, but not around any specific posture in the walker stimuli. These findings suggest that there is a low-level “punch detector”, that, we show, is also view-point independent. This detector is action category specific, and is similar to, but different from the “life detector” postulated in previous research. It helps humans identify aggressive punching behavior in a crowd. We tentatively suggest that this mechanism operates through a subcortical “threat”-detection mechanism.
This research was supported by NSF grant BCS-0843880.