Hierarchical structure is employed by humans during visual motion perception

Abstract

In the real world, complex dynamic scenes often arise from the composition of simpler parts. The visual system exploits this structure by hierarchically decomposing dynamic scenes: When we see a person walking on a train or an animal running in a herd, we recognize the individual’s movement as nested within a reference frame that is, itself, moving. Despite its ubiquity, surprisingly little is understood about the computations underlying hierarchical motion perception. To address this gap, we developed a class of stimuli that grant tight control over statistical relations among object velocities in dynamic scenes. We first demonstrate that structured motion stimuli benefit human multiple object tracking performance. Computational analysis revealed that the performance gain is best explained by human participants making use of motion relations during tracking. A second experiment, using a motion prediction task, reinforced this conclusion and provided fine-grained information about how the visual system flexibly exploits motion structure.

Human visual motion perception shows hallmarks of Bayesian structural inference

“…The observed stochasticity of human responses to identical scenes (trial repetitions) is in line with previous studies 16,19,20,32 , which we were able to partially capture with generalized probability matching, by raising the . For most participants, the full model explains human responses better than models with a uniform prior (all b S = 0 , "bias-free"), without lapses ( π L = 0 , "lapse-free"), and a model without the Bayesian core, p(X | S) , of the Kalman filter ("non-Bayesian").…”

“…Visual motion scenes with hierarchical structure. We adopted the hierarchical motion structure representation from Gershman et al 6 and Bill et al 16 , and the stimulus generation from Bill et al 16 . Their main components with respect to the present work are summarized in the following.…”

“…1A do not specify how a motion structure gives rise to a concrete dynamic visual scene. We created motion scenes for our experiments using the stochastic stimulus generation proposed in 16 (see Materials and Methods for details). In this stimulus design, dot velocities…”

“…For example, the global motion source has zero strength in the clustered (C) condition. Bill et al 16 proposed that the hierarchical motion structures S from Stochastic generation of visual scenes. Dot velocities were generated stochastically from a multivariate Ornstein-Uhlenbeck process.…”

“…Recently, Gershman et al 6 proposed a Bayesian framework for explaining motion structure discovery, using probabilistic inference over hierarchical motion structures (they called motion trees ). Subsequently, Bill et al 16 established experimentally that humans can exploit hierarchical structure when solving demanding visual tasks, such as object tracking and trajectory prediction. However, in this study, participants were informed about the structure underlying each trial, and the tested motion relations were designed to be clearly distinguishable.…”