Coherent Motion Patterns Research Articles

Embedding group and obstacle information in LSTM networks for human trajectory prediction in crowded scenes

Recurrent neural networks have shown good abilities in learning the spatio-temporal dependencies of moving agents in crowded scenes. Recently, they have been adopted to predict the motion of pedestrians by learning the relative motion of each individual in the crowd with respect to its neighbors. Crowded scenes present a wide variety of situations, which do not depend solely on the agents’ positions, but also relate to the structure of the environment, the density of the crowd, and the social relationships between pedestrians. In this work we propose a framework to improve the state-of-the-art models of crowd motion prediction by enriching the learning model with the social relationships between pedestrians walking in the crowd, as well as the layout of the environment. We observe that socially-related people tend to exhibit coherent motion patterns. Exploiting the motion coherency, we are able to cluster trajectories with similar motion properties and improve the trajectory prediction, especially at the group level. Furthermore, we incorporate into the model also the layout of the environment, to guarantee a more realistic and reliable learning framework. We evaluate our approach on standard crowd benchmark datasets, demonstrating its efficacy and applicability, improving the accuracy in trajectory prediction.

Exploring Coherent Motion Patterns via Structured Trajectory Learning for Crowd Mood Modeling

Crowd behavior analysis has recently attracted extensive attention in research. However, the existing research mainly focuses on investigating motion patterns in crowds, while the emotional aspects of crowd behaviors are left unexplored. Analyzing the emotion of crowd behaviors is indeed extremely important, as it uncovers the social moods that are beneficial for video surveillance. In this paper, we propose a novel crowd representation termed crowd mood. Crowd mood is established based upon the discovery that the social emotional hypothesis of crowd behaviors can be revealed by investigating the spacing interactions and the structural levels of motion patterns in crowds. To this end, we first learn the structured trajectories of crowds by particle advection using low-rank approximation with group sparsity constraint, which implicitly characterizes the coherent motion patterns. Second, rich emotional motion features are explicitly extracted and fused by support vector regression to reflect social characteristics. In particular, we construct weighted features in a boosted manner by considering the features’ significance. Finally, crowd mood is intuitively presented as affective curves to track the emotion states of the crowd dynamics, which is robust to noise, sensitive to semantic shift, and compact for pattern expressions. Extensive evaluations on crowd video data sets demonstrate that our approach effectively models crowd mood and achieves significantly better results with comparisons to several alternative and state-of-the-art approaches for various tasks, i.e., crowd mood classification, global abnormal mood detection, and crowd emotion matching.

A neural model of the temporal dynamics of figure–ground segregation in motion perception

How does the visual system manage to segment a visual scene into surfaces and objects and manage to attend to a target object? Based on psychological and physiological investigations, it has been proposed that the perceptual organization and segmentation of a scene is achieved by the processing at different levels of the visual cortical hierarchy. According to this, motion onset detection, motion-defined shape segregation, and target selection are accomplished by processes which bind together simple features into fragments of increasingly complex configurations at different levels in the processing hierarchy. As an alternative to this hierarchical processing hypothesis, it has been proposed that the processing stages for feature detection and segregation are reflected in different temporal episodes in the response patterns of individual neurons. Such temporal epochs have been observed in the activation pattern of neurons as low as in area V1. Here, we present a neural network model of motion detection, figure-ground segregation and attentive selection which explains these response patterns in an unifying framework. Based on known principles of functional architecture of the visual cortex, we propose that initial motion and motion boundaries are detected at different and hierarchically organized stages in the dorsal pathway. Visual shapes that are defined by boundaries, which were generated from juxtaposed opponent motions, are represented at different stages in the ventral pathway. Model areas in the different pathways interact through feedforward and modulating feedback, while mutual interactions enable the communication between motion and form representations. Selective attention is devoted to shape representations by sending modulating feedback signals from higher levels (working memory) to intermediate levels to enhance their responses. Areas in the motion and form pathway are coupled through top-down feedback with V1 cells at the bottom end of the hierarchy. We propose that the different temporal episodes in the response pattern of V1 cells, as recorded in recent experiments, reflect the strength of modulating feedback signals. This feedback results from the consolidated shape representations from coherent motion patterns and the attentive modulation of responses along the cortical hierarchy. The model makes testable predictions concerning the duration and delay of the temporal episodes of V1 cell responses as well as their response variations that were caused by modulating feedback signals.

Quantitative object motion prediction by an ART2 and Madaline combined neural network: Concepts and experiments

A neural network that combines the structures of an ART2 network and a Madaline network is applied to the prediction of object motions in dynamic environments. By the use of an ART2 network, the system extracts a set of coherent motion patterns from a sequence of noise-corrupted input signals. This process is done in an unsupervised learning mode. The system then uses the extracted patterns to direct the quantitative prediction of the future motion states of the object through a Madaline network. A push-forward shift-register keeps track of the object motion profile in real time, and provides input signals to both networks. The method does not require any assumption of mathematical models for the object motions, and is applicable to a variety of situations, such as environments with cooperating robots, where the object motions exhibit quite irregular and vigorous variations.

Coherent Compound Motion: Corners and Nonrigid Configurations

Consider two wire gratings, superimposed and moving across each other. Under certain conditions the two gratings will cohere into a single, compound pattern, which will appear to be moving in another direction. Such coherent motion patterns have been studied for sinusoidal component gratings, and give rise to percepts of rigid, planar motions. In this paper we show how to construct coherent motion displays that give rise to nonuniform, nonrigid, and nonplanar percepts. Most significantly, they also can define percepts with corners. Since these patterns are more consistent with the structure of natural scenes than rigid sinusoidal gratings, they stand as interesting stimuli for both computational and physiological studies. To illustrate, our display with sharp corners (tangent discontinuities or singularities) separating regions of coherent motion suggests that smoothing does not cross tangent discontinuities, a point that argues against existing (regularization) algorithms for computing motion. This leads us to consider how singularities can be confronted directly within optical flow computations, and we conclude with two hypotheses: (1) that singularities are represented within the motion system as multiple directions at the same retinotopic location; and (2) for component gratings to cohere, they must be at the same depth from the viewer. Both hypotheses have implications for the neural computation of coherent motion.