We can recognize the faces of our friends very quickly from just a snapshot. Within 150 milliseconds of being flashed a photo, brain signals respond differently to photos containing animals than photos with no animals. We can categorize scenes as “beach,” “forest,” or “city” when they are flashed for even shorter periods.
But we also get a great deal of information from the motion of people and animals. We can identify our friends and family members just from a point-light display of them walking. We can also detect the emotions of point-light faces, and even the species of point-light animals.
Fascinating as point-light displays are, however, we rarely see them in real life. Point-light displays suggest that motion gives us a great deal of information about the object we’re looking at, but we can’t be sure that real-world perception works the same way. A team led by Quoc Vuong has conducted a study to see if what we know about point-light displays transfers to real-world objects and scenes.
They constructed a set of composite movies like this one (QuickTime required):
In every clip, a machine was superimposed with either a walking human figure (like in this clip) or another machine. To make the task even more difficult, viewers were shown two movies simultaneously, for just two thirds of a second. Viewers had to determine if one of the two movies included a human. The relative visibility of the human figure was also varied.
Even more critically, half of the images they saw were animated, and half were still photos. Here are the results:
As you’d expect, the more visible the human, the higher the accuracy. But no matter the visibility level of the human figure, viewers were more accurate identifying humans when in the animated sequence compared to still photos.
One possible objection to these results is that the animated sequences are easier to process, whether or not they contain a human. Vuong’s team repeated the experiment, but instead of the completely still shots, they used movies of animated machines superimposed with still humans. Here are the results:
Viewers were better at identifying the still humans with animated machines, but they were still significantly better at spotting humans when both the machine and human were animated. These results also held when the experiment was repeated with upside-down movies, and with animals instead of machines, like in this movie:
So our impressive ability to identify point-light displays extends to more natural-looking movies. Vuong’s team believes that the upside-down experiment is an indicator that their results may apply to a range of different objects in motion, such as animals and even machines. If we can detect upside-down walking humans better than upside-down still humans, we’re probably also better at detecting walking mice, cats, or goats.
Vuong, Q.C., Hof, A.F., Bülthoff, H.H., & Thornton, I.M. (2006). An advantage for detecting dynamic targets in natural scenes. Journal of Vision, 6, 87-96.