Abstract not found

We present a biologically-motivated system for the recognition of actions from video sequences. The approach builds on recent work on object recognition based on hierarchical feedforward architectures [25, 16, 20] and extends a neurobiological model of motion processing in the visual cortex [10]. The system consists of a hierarchy of spatio-temporal feature detectors of increasing complexity: an input sequence is first analyzed by an array of motiondirection sensitive units which, through a hierarchy of processing stages, lead to position-invariant spatio-temporal feature detectors. We experiment with different types of motion-direction sensitive units as well as different system architectures. As in [16], we find that sparse features in intermediate stages outperform dense ones and that using a simple feature selection approach leads to an efficient system that performs better with far fewer features. We test the approach on different publicly available action datasets, in all cases achieving the highest results reported to date. 1.

Abstract not found

Six experiments demonstrate a visual dynamic illusion. Previous work has shown that in 2-dimensional (2D) drawing movements, tangential velocity and radius of curvature covary in a constrained manner. The velocity of point stimuli is perceived as uniform if and only if this biological constraint is satisfied. The illusion is conspicuous: The variations of velocity in the stimuli exceed 200%. Yet movements are perceived as uniform. Conversely, 2D stimuli moving at constant velocity are perceived as strongly nonuniform. The illusion is robust: Exposure to true constant velocity fails to suppress it. Results cannot be explained entirely by the kinetic depth effect. The illusion is evidence of a coupling between motor and perceptual processes: Even in the absence of any intention to perform a movement, certain properties of the motor system implicitly influence perceptual interpretation of the visual stimulus. Many experimental facts can support the general view that visual percepts arise less as a direct, inescapable consequence of the sensory data than as the result of a selection process that is based on sensory cues and acts on a set of alternatives available a priori to the perceptual system. This view can be

Physical imagery occurs when people imagine one object causing a change to a second object. To make inferences through physical imagery, people must represent information that coordinates the interactions among the imagined objects. The current research contrasts two proposals for how this coordinating information is realized in physical imagery. In the traditional kinematic formulation, imagery transformations are coordinated by geometric information in analog spatial representations. In the dynamic formulation, transformations may also be regulated by analog representations of force and resistance. Four experiments support the dynamic formulation. They show, for example, that without making changes to the spatial properties of a problem, dynamic perceptual information (e.g., torque) and beliefs about physical properties (e.g., viscosity) affect the inferences that people draw through imagery. The studies suggest that physical imagery is not so much an analog of visual perception as it is an analog of physical action. A simple model that represents force as a rate helps explain why inferences can emerge through imagined actions even though people may not know the answer explicitly. It also explains how and

In classic demonstrations of apparent motion, observers typically report seeing motion along the shortest possible path between 2 sequentially presented objects. However, when realistic photographs of a human body are sequentially presented at slow temporal rates, observers report paths of apparent motion that are consistent with the movement limitations of the human body even when those paths are not the shortest possible. The current set of experiments examined those aspects of the human form that lead to the perception of biomechanically consistent paths of motion. The authors ' findings suggest that the perception of apparent biological motion extends to human movements that involve inanimate objects. The authors also report that observers can perceive apparent motion of nonbiological objects in a manner similar to apparent motion of human bodies. However, a global hierarchy of orientation and position cues resembling the human form is required for the perception of these paths. Our successful interaction with the environment depends largely on our ability to interpret visual forms and motions accurately and rapidly. This ability appears to be particularly important for the visual perception of human movement. Johansson and his colleagues generated extensive interest in the perception of human movement with the demonstration that even under extremely reduced conditions, observers could readily identify human locomotion

More than 50 years after the appearance of the motor theory of speech perception, it is timely to evaluate its three main claims that (1) speech processing is special, (2) perceiving speech is perceiving gestures, and (3) the motor system is recruited for perceiving speech. We argue that to the extent that it can be evaluated, the first claim is likely false. As for the second claim, we review findings that support it and argue that although each of these findings may be explained by alternative accounts, the claim provides a single coherent account. As for the third claim, we review findings in the literature that support it at different levels of generality and argue that the claim anticipated a theme that has become widespread in cognitive science.

In two experiments, perceptual anticipation—that is, the observer’s ability to predict the course of dynamic visual events—in the case of handwriting traces was investigated. Observers were shown the dynamic display of the middle letter l excerpted from two cursive trigrams (lll or lln) handwritten by one individual. The experimental factor was the distribution of the velocity along the trace, which was controlled by a single parameter, β. Only for one value of this parameter (β = 2/3) did the display comply with the two-thirds power law, which describes how tangential velocity depends on curvature in writing movements. The task was to indicate the trigram from which the trace was excerpted—that is, to guess the letter that followed the specific instance of the l that had been displayed. In Experiment 1, the no answer option was available. Experiment 2 adopted a forced-choice response rule. Responses were never reinforced. When β = 2/3, the rate of correct guesses was high (Experiment 1, P{correct} =.69; Experiment 2, P{correct} =.78). The probability of a correct answer decreased significantly for both smaller and larger values of β, with wrong answers becoming predominant at the extremes of the range of variation of this parameter. The results are consistent with the hypothesis that perceptual anticipation of human movements involves comparing the perceptual stimulus with an internal dynamic representation of the ongoing event. Biological motion—that is, voluntary movements of the body—has peculiar perceptual qualities acknowledged and investigated since the beginning of the century (Kenkel, 1913; Korte, 1915; Wertheimer, 1912). Salience is one striking feature of biological motion: Even very sketchy descriptions of whole-body movements can be detected (Johansson, 1950) and discriminated (Beardsworth & Buckner, 1981) accurately, within a few hundred

Abstract not found

Transformational apparent motion (TAM) occurs when a figure changes discretely from one configuration to another overlapping configuration. Rather than an abrupt shape change, the initial shape is perceived to transform smoothly into the final shape as if animated by a series of intermediate shapes. We find that TAM follows an analysis of form that takes 80–140 msec. Form analysis can function both at and away from equiluminance and can occur over contours defined by uniform regions as well as outlines. Moreover, the forms analyzed can be 3-D, resulting in motion paths that appear to smoothly project out from or into the stimulus plane. The perceived transformation is generally the one that involves the least change in the shape or location of the initial figure in a 3-D sense. We conclude that perception of TAM follows an analysis of 3-D form that takes,100 msec. This stage of form analysis may be common to both TAM and second-order motion. When two nonoverlapping figures are flickered in succession within a certain range of spatiotemporal offsets (Korte, 1915), they appear to comprise a single object jumping rigidly back and forth in translational apparent motion. Because no object actually moves in the world, the