Learning behavior patterns from video for agent-based crowd modeling and simulation

This paper proposes a novel data-driven modeling framework to construct agent-based crowd model based on real-world video data. The constructed crowd model can generate crowd behaviors that match those observed in the video and can be used to predict trajectories of pedestrians in the same scenario. In the proposed framework, a dual-layer architecture is proposed to model crowd behaviors. The bottom layer models the microscopic collision avoidance behaviors, while the top layer models the macroscopic crowd behaviors such as the goal selection patterns and the path navigation patterns. An automatic learning algorithm is proposed to learn behavior patterns from video data. The learned behavior patterns are then integrated into the dual-layer architecture to generate realistic crowd behaviors. To validate its effectiveness, the proposed framework is applied to two different real world scenarios. The simulation results demonstrate that the proposed framework can generate crowd behaviors similar to those observed in the videos in terms of crowd density distribution. In addition, the proposed framework can also offer promising performance on predicting the trajectories of pedestrians.     

数据驱动的汽车总装线建模与仿真技术研究

现代制造业的生产模式向柔性制造和客户化定制方向发展,为了提高制造系统的快速响应能力,实现快速仿真分析,提出基于数据驱动的建模与仿真方法,其具有建模周期短、模型可读性、可重用性较高的特点.利用IDEFIX建模方法建立模型的底层数据结构,基于Arena构建仿真模型生成器,利用已建好的数据模型提供的属性信息,自动生成目标系统...     

基于流形学习和混合模型的视频摘要方法

视频摘要是进行视频浏览、视频检索、视频索引等视频应用的前提,而且视频摘要类似于文本的摘要,也是对视频内容的一个简短概括。为了自动获得既包含视频的主要信息,而冗余信息又少的视频摘要,提出了一种基于流形学习和有限混合模型的自动视频摘要方法。该方法通过对视频序列进行流形建模,首先得到视频场景的初次分割;然后对包含内容较多的场景,使用等距降维方法计算视频帧的特征向量;最后将视频帧的特征向量输入到混合模型进行聚类分析,得到更细粒度的摘要结果。为了实现视频摘要的自动处理,所采用的混合模型需要具有模型选择功能。混合模型的聚类结果和流形建模的结果共同构成了视频摘要。视频分割片段的实验结果表明,在不需人为干预的情况下,所提供的视频摘要不仅包含视频主要内容,而且冗余信息少。     

Crowds by Example

We present an example-based crowd simulation technique. Most crowd simulation techniques assume that the behavior exhibited by each person in the crowd can be defined by a restricted set of rules. This assumption limits the behavioral complexity of the simulated agents. By learning from real-world examples, our autonomous agents display complex natural behaviors that are often missing in crowd simulations. Examples are created from tracked video segments of real pedestrian crowds. During a simulation, autonomous agents search for examples that closely match the situation that they are facing. Trajectories taken by real people in similar situations, are copied to the simulated agents, resulting in seemingly natural behaviors.     

Automating test generation for discrete event oriented embedded systems

A method for the automatic generation of test scenarios from the behavioral requirements of a system is presented in this paper. The generated suite of test scenarios validates the system design or implementation against the requirements. The approach proposed here uses a requirements model and a set of four algorithms. The requirements model is an executable model of the proposed system defined in a deterministic state-based modeling formalism. Each action in the requirements model that changes the state of the model is identified with a unique requirement identifier. The scenario generation algorithms perform controlled simulations of the requirements model in order to generate a suite of test scenarios applicable for black box testing. Measurements of several metrics on the scenario generation algorithms have been collected using prototype tools.     

基于多尺度卷积神经网络的人群聚集异常预测

已有的公共场所人群聚集异常行为检测方法较少,且大多检测方法都是在人群已经异常聚集后再进行检测,检测准确率不高,时效性不够好。提出一种基于多尺度卷积神经网络（MCNN）的人群聚集异常预测模型。首先,通过多尺度卷积神经网络训练一个人群计数模型,用训练好的模型对人群聚集异常视频进行测试;然后在测试中完成人群人数统计与人群头部坐标点获取,进而计算人群密度、人群距离势能与人群分布熵;最后将得到的3种人群运动状态特征值利用PSO-ELM进行训练,得到预测模型,通过特征数据的变化,完成人群聚集行为的预测。实验结果表明,与现有算法相比,该模型能有效实现人群聚集异常行为的预警与检测,时效性强,为采取相应应急措施提供了更多时间,预测准确率达到了9717%。     

Crowd simulation for emergency response using BDI agents based on immersive virtual reality

This paper presents a novel methodology involving a Virtual Reality (VR)-based Belief, Desire, and Intention (BDI) software agent to construct crowd simulation and demonstrates the use of the same for crowd evacuation management under terrorist bomb attacks in public areas. The proposed BDI agent framework allows modeling of human behavior with a high degree of fidelity. The realistic attributes that govern the BDI characteristics of the agent are reverse-engineered by conducting human-in-the-loop experiments in the VR-based Cave Automatic Virtual Environment (CAVE). To enhance generality and interoperability of the proposed crowd simulation modeling scheme, input data models have been developed to define environment attributes (e.g., maps, demographics, evacuation management parameters). The validity of the proposed data models are tested with two different evacuation scenarios. Finally, experiments are conducted to demonstrate the effect of various crowd evacuation management parameters on the key performance indicators in the evacuation scenario such as crowd evacuation rate and densities. The results reveal that constructed simulation can be used as an effective emergency management tool.     

Modeling Groups of Plausible Virtual Pedestrians

In a proposed methodology for modeling dynamic crowd scenarios, a video corpus informs the modeling process, after which the resultant animations undergo perception-based evaluation. The aim is to improve the crowd's visual plausibility rather than the simulation's correctness. A real-life crowd animation system demonstrates the methodology's practical application.     

A Data‐Driven Framework for Visual Crowd Analysis

We present a novel approach for analyzing the quality of multi‐agent crowd simulation algorithms. Our approach is data‐driven, taking as input a set of user‐defined metrics and reference training data, either synthetic or from video footage of real crowds. Given a simulation, we formulate the crowd analysis problem as an anomaly detection problem and exploit state‐of‐the‐art outlier detection algorithms to address it. To that end, we introduce a new framework for the visual analysis of crowd simulations. Our framework allows us to capture potentially erroneous behaviors on a per‐agent basis either by automatically detecting outliers based on individual evaluation metrics or by accounting for multiple evaluation criteria in a principled fashion using Principle Component Analysis and the notion of Pareto Optimality. We discuss optimizations necessary to allow real‐time performance on large datasets and demonstrate the applicability of our framework through the analysis of simulations created by several widely‐used methods, including a simulation from a commercial game.     

Rotary drying process modeling and online compensation

Rotary drying process modeling is a complex procedure due to the difficulties in measurement and estimation of kinetic model parameters. To solve the problem, a hybrid modeling method with online compensation is proposed in the present study. A mathematical model is built to describe the axial characteristics of rotary drying process. The drying rate which is the key parameter in the model is estimated by using a SVR-based fuzzy modeling approach, which can automatically extract fuzzy IF-THEN rules from support vectors. Laboratory experiments are conducted to obtain the drying rate sample data for the modeling purpose. In order to reduce the modeling errors for an industrial rotary dryer and improve the hybrid model prediction accuracy, an online matching coefficient is introduced, and a method based on improved online SVR is then applied for modeling error compensation. The experiment dada based modeling results have verified the effectiveness and demonstrated the accuracy and adaptability of the proposed hybrid modeling method.     

人群应急疏散可视仿真研究进展和问题

人群应急疏散可视仿真是用智能体来模拟具有自主感知、情绪和行为能力的人群个体,并采用3维可视的方式来直观呈现人群应急疏散情景,可以为制定人群应急预案提供形象直观的分析方法。本文从人群仿真数据的来源、人群导航模型的构建、人群行为模型、人群情绪感染、人群渲染5个方面概述目前研究的进展,然后从仿真模型的可验证性、人群疏散导航模型的构建、人与环境的物理模型、动物逃生实验与仿真、疏散中的社会行为表现以及人群情绪的可视计算6个角度讨论需要进一步研究的问题。针对需要深入研究的问题,指出借助于紧急事件的视频监控分析和虚拟人群情景的用户调查,有助于完善人群仿真模型。结合物理模型,可以更准确地描述人群应急疏散场景。开展动物逃生实验分析,有助于完善人群运动导航算法。建立人群社会行为模型,可以更详细描述疏散中人群行为的多样性。构建基于多通道感知的人群情绪感染计算方法,可以详尽描述情绪感染的过程。人群应急疏散行为的可视仿真研究在城市的安全管理方面具有重要的应用前景,但其研究仍存在很多亟待解决的问题,综合地运用多学科知识,完善实验手段是进一步推动研究的关键所在。     

一种基于视频数据的服装显著性预测方法

人眼视觉注意机制表明当人眼观察目标时,注意力只会放在少数感兴趣的区域, 而自动忽略视野中大部分不感兴趣的其他区域。研究人类视觉注意机制,并构建有效的服装显 著性预测模型,可在后期用于指导更加逼真有效的服装运动建模,提高模拟效率。为此,对着 装人体运动视频数据进行分析,构造了种类多样的视频样本,并利用眼动技术采集真实人眼的 注视数据,采用高斯卷积生成视频帧的显著图作为训练模型所需的 Ground-truth。在进行视频特 征提取时,结合了底层图像特征、高层语义特征以及运动特征,共同构造特征向量和标签,并 通过支持向量机(SVM)训练得到基于服装视频的显著性预测模型。通过实验验证,该方法的性 能在服装显著性预测时,优于传统的显著性预测算法,具有一定的鲁棒性。     

多输入多输出系统的数据分层建模

为了高效准确地实现多输入多输出系统的数据建模,本文提出了数据分层建模算法.将多输入多输出数据建模问题分解为一系列单输入多输出的建模问题,同时改进标准遗传编程算法,将单树表示的的个体结构扩展为向量树的进化个体,实现多输出系统的自动建模.通过大量的计算实例表明,这种算法可以实现复杂的多输入多输出系统的建模,提高数据建模的效率和精度.     

Formal design and implementation of constraints in software components

This paper deals with the modeling, automatic implementation and runtime verification of constraints in component-based applications. Constraints have been assuming an ever more relevant role in modeling distributed systems as long as business rules implementation, design-by-contract practice, and fault-tolerance requirements are concerned. Nevertheless, component developers are not sufficiently supported by existing tools to model and implement such features.In this paper, we propose a methodology and a set of tools that enable developers both to model component constraints and to generate automatically component skeletons that already implement such constraints. The methodology has been extended to support implementation even in case of legacy components.     

Hierarchical crowd analysis and anomaly detection

ObjectiveThis work proposes a novel approach to model the spatiotemporal distribution of crowd motions and detect anomalous events.MethodsWe first learn the regions of interest (ROIs) which inform the behavioral patterns by trajectory analysis with Hierarchical Dirichlet Processes (HDP), so that the main trends of crowd motions can be modeled. Based on the ROIs, we then build a series of histograms both on global and local levels as the templates for the observed movement distribution, which statistically describes time-correlated crowd events. Once the template has been built hierarchically, we import real data containing the discrete trajectory observations from video surveillance and detect abnormal events for individuals and for crowds.ResultsExperimental results show the effectiveness of our approach, which is able to analyze and extract the crowd motion information from observed trajectory dataset, and achieve the anomaly detection at the hierarchical levels.ConclusionThe proposed hierarchical approach can learn the moving trends of crowd both in global and local area and describe the crowd behaviors in statistical way, which build a template for pedestrian movement distribution that allows for the detection of time-correlated abnormal crowd events.     

Optimal crowd editing

Simulating realistic crowd behaviors is a challenging problem in computer graphics. Yet, several satisfying simulation models exhibiting natural pedestrians or group emerging behaviors exist. Choosing among these model generally depends on the considered crowd density or the topology of the environment. Conversely, achieving a user-desired kinematic or dynamic pattern at a given instant of the simulation reveals to be much more tedious. In this paper, a novel generic control methodology is proposed to solve this crowd editing issue. Our method relies on an adjoint formulation of the underlying optimization procedure. It is independent to a certain extent of the choice of the simulation model, and is designed to handle several forms of constraints. A variety of examples attesting the benefits of our approach are proposed, along with quantitative performance measures.     

Velocity-based modeling of physical interactions in dense crowds

We present an interactive algorithm to model physics-based interactions in dense crowds. Our approach is capable of modeling both physical forces and interactions between agents and obstacles, while also allowing the agents to anticipate and avoid upcoming collisions during local navigation. We combine velocity-based collision-avoidance algorithms with external physical forces. The overall formulation produces various effects of forces acting on agents and crowds, including balance recovery motion and force propagation through the crowd. We further extend our method to model more complex behaviors involving social and cultural rules. We use finite-state machines to specify a series of behaviors and demonstrate our approach on many complex scenarios. Our algorithm can simulate a few thousand agents at interactive rates and can generate many emergent behaviors.