Hierarchical structure is employed by humans during visual motion perception

The structured organization of motion in visual scenes is highly informative for our everyday perception: We recognize people by the way they walk, track objects through occlusion, or predict hazardous situations from the traffic flow. It is, however, unclear how our minds tame the overwhelmingly complex stream of dynamic information received by the retina to form such stable percepts. We argue that an observer can exploit a "divide-and-conquer" strategy where complex motion relations are broken down into compositions of simpler motions. Evidence for hierarchical decomposition comes from multiple object tracking and prediction experiments in which humans are able to exploit motion structure knowledge to improve performance. Our results can guide neuroscience experiments on the neural representation of structure.