位置：立项数据库 > 立项详情页
基于多摄像头协同的运动对象检测跟踪和异常行为分析
  • 项目名称：基于多摄像头协同的运动对象检测跟踪和异常行为分析
  • 项目类别：重点项目
  • 批准号：61035001
  • 申请代码：F0304
  • 项目来源：国家自然科学基金
  • 研究期限：2011-01-01-2014-12-31
  • 项目负责人：田永鸿
  • 负责人职称：副教授
  • 依托单位：北京大学
  • 批准年度：2010
中文摘要：

视频监控系统已经成为敏感场所安全防范、安全隐患发现和应急管理的重要基础设施。然而，花费巨额资金建立的视频监控网络并未发挥应有效果，瓶颈在于缺乏有效的自动分析识别技术来高效处理大量的实时视频数据。本项目以大规模多摄像机监控视频为对象，突破多视场景建模、多视环境下的对象检测和协同跟踪、异常行为检测与识别等关键技术，提出多视角监控视频中运动对象关联分析模型与方法和适合异常行为模式挖掘的机器学习方法，建立一套先进的多摄像机协同的运动对象检测、跟踪与异常行为分析理论和技术体系。本项目还将通过开发应用测试验证系统、参加TRECVID监控视频事件检测国际比赛和下一代监控视频编码国际标准与国家标准来验证所提理论、方法与技术的有效性，为新一代智能视频监控系统提供核心技术。

中文主题词： 多视贝叶斯网络；序列判别学习；社会力属性模型；协同变量Boost方法；城市交通异常事件检测平台
英文摘要：

Multi-view Bayesian network；Sequence discriminative learning；Social attribute-aware force model；CovBoost；Metro-transport abnormal event detection platform

英文主题词： Multi-view Bayesian network；Sequence discriminative learning；Social attribute-aware force model；CovBoost；Metro-transport abnormal event detection platform
结论摘要：

本项目围绕多摄像头协同的运动对象检测跟踪和异常行为分析这一科学问题，从模型方法、关键技术、应用验证三个层面开展研究工作。主要技术创新如下 (1)跨摄像头对象检测跟踪理论与方法针对多摄像头协同的对象检测跟踪问题，提出一整套运动对象关联分析模型、对象表示方法、跨摄像头对象检测与跟踪技术，包括多视贝叶斯网络模型MvBN、基于混合范式正则化的多义性视觉表示方法、基于迁移学习的对象检测与跟踪方法、基于MvBN和高度自适应映射的多视对象检测方法、基于表观和步态的行人再标识方法等，获IEEE多摄像机对象检测与跟踪评测PETS2012中多项性能指标的第一，并被评为近五年PETS评测中性能最好的算法。 (2)基于学习的监控视频行为与异常事件检测方法从学习方法着手，提出了一系列监控视频行为分析和异常检测方法，包括基于深度轨迹描述子的动作表示方法、面向群体行为描述的轨迹图包方法、面向近似视频对齐的多粒度子序列匹配方法、融合运动和表观信息的多人群组行为识别方法、基于序列判别学习的预定义监控事件检测方法、基于模糊聚类多自编码器的无监督异常发现方法、基于社会属性力的群体行为异常检测方法等，在TRECVID监控视频事件检测比赛中连续2011～2012两年均在多项检测任务上性能最优。 (3)监控视频前背景分析技术及其在标准中的贡献。针对前背景建模与分析问题，提出了多项高性能的分析方法，包括像素级选择式特征背景建模方法、基于投影模型的自适应运动阴影检测方法、基于大规模统计先验和多任务排序学习的视觉显著计算方法、基于互补显著先验的前景对象分割方法、簇敏感多核学习方法GS-MKL，获2013年ImageCLEF Robot Vision竞赛冠军。将场景分析嵌入编码框架，提出了基于背景建模的监控视频编码及优化技术，实现了压缩效率翻番，已作为IEEE国际标准颁布。项目构建了目前国内外最大规模的监控视频对象检测与跟踪数据集PKU-SVD-B，实现了多摄像头协同对象检测与追踪原型系统，开发了城市交通异常事件检测平台并在青岛30多处应用。课题组出版专著1部，待出版1部；发表学术论文73篇，其中国际期刊27篇、CVPR/ICCV等顶级会议4篇；申请发明专利25项，其中2项已授权，获软件著作权5项；培养博士后4人，博士生8人，硕士生22人。因此，项目圆满完成任务，各项指标均已大幅超过了任务书的规定。

成果综合统计
成果类型
数量
  • 期刊论文
  • 会议论文
  • 专利
  • 获奖
  • 著作
  • 38
  • 42
  • 10
  • 0
  • 2
期刊论文
车载视觉系统中的行人检测技术综述
Content-Based Copy Detection through Multimodal Feature Representation and Temporal Pyramid Matching
Visual Saliency with Statistical Priors
IEEE 1857 Standard: Empowering Smart Video Surveillance Systems
Representing Visual Objects in HEVC Coding Loop
SSOCBT: A Robust Semi-Supervised Online CovBoost Tracker by Using Samples Differently
Multi-Task Rank Learning for Visual Saliency Estimation
Transferring Boosted Detectors Towards Viewpoint and Scene Adaptiveness
Learning Complementary Saliency Priors for Foreground Object Segmentation in Complex Scenes
Detection and Localization with Scalable Cascading of Complementary Detectors and Multi-scale Sequen
Video Copy-Detection and Localization with a Scalable Cascading Framework
IEEE 1857: Boosting Video Applications in CPSS
Weighted Visual Vocabulary to Balance the Descriptive Ability on General Dataset
INSTRE: a New Benchmark for Instance-Level Object Retrieval and Recognition
Representing dense crowd patterns using bag of trajectory graphs
Optimizing the Hierarchical Prediction and Coding in HEVC for Surveillance and Conference Videos wit
Scalable storyboards in handheld devices: applications and evaluation metrics
Multi-Task Rank Learning for Visual Saliency in Video
Selective Eigenbackground for Background Modeling and Subtraction in Crowded Scenes
TASC: A Transformation-Aware Soft Cascading Approach for Multimodal Video Copy Detection
Foreground Object Segmentation in Complex Scenes with Complementary Saliency Priors
Recognizing Human Group Action by Layered Model with Multiple Cues
Learning Hierarchical Semantic Description via Mixed-norm Regularization for Image Understanding
Robust Multiple Cameras Pedestrian Detection with Multi-view Bayesian Network
Social Attribute-aware Force Model: Exploiting Richness of Interaction for Abnormal Crowd Detection
Relative Image Similarity Learning with Contextual Information for Internet Cross-media Retrieval
Fast and Efficient Transcoding Based on Low-complexity Background Modeling and Adaptive Block Classi
特征提取和匹配的图像倾斜校正
基于视觉的人体动作识别综述
会议论文
Overview of the IEEE 1857 Surveillance Groups
PKU-NEC @ TRECVID 2011 SED: Sequence-Based Event Detection in Surveillance Video
MPLBoost-based Mixture Model for Effective Human Detection with Deformable Part Model
A Background Proportion Adaptive Lagrange Multiple Selection Method for Surveillance Video on High E
MIAR ICT participation at Robot Vision 2013
Multi-view Gait Recognition with Incompelete Training Data
ObjectSense: a scalable multi-objects recognition system based on partial-duplicate image retrieval
<span style="font-family:Calibri, sans-serif;font-size:14px;">Selective Eigenbackgro
Treat Samples Differently: Object Tracking with Semi-supervised Online CovBoost
Human Group Activity Analysis with Fusion of Motion and Appearance Information
Human Tracking by Structured Body Parts
Justifying the Importance of Color Cues in Object Detection: a Case Study on Pedestrian
Background-Foreground Division based Search for Motion Estimation in Surveillance Video Coding
Learning sparse prototypes for crowd perception via Ensemble Coding Mechanisms
Accuracy and specificity trade-off in k-nearest neighbors classification
Color Maximal-Dissimilarity Pattern for Pedestrian Detection
Set-based classification for person re-identification utilizing mutual-information
Learning Image Vicept Description via Mixed-Norm Regularization for Large Scale Semantic Image Searc
An Efficient Background Reconstruction Based Coding Method for Surveillance Videos Captured By Movin
Swiss-System based Cascade Ranking for Gait-based Person Re-identification
Surveillance Video Coding with Quadtree Partition Based ROI Extraction
Recognizing Realistic Action Using Contextual Feature Group
Abnormal Crowd Behavior Detection Based on Social Attribute-Aware Force Model
Multi-camera Pedestrian Detection with a Multi-view Bayesian Network Model
PKU-NEC @ TRECVid 2012 SED: Uneven-Sequence Based Event Detection in Surveillance Video
Video Picture-in-Picture Detection using Spatio-Temporal Slicing
A Coding Unit Classification Based AVC-to-HEVC Transcoding with Background Modeling for Surveillance
Wavelet Based Smoke Detection Method with RGB Constrast-Image and Shape Constraint
Pair-wise Event Detection using Cubic Features and Sequence Discriminant Learning
A System based on Sequence Learning for Event Detection in Surveillance Video
Hierarchical-and-Adaptive Bit-allocation with Selective Background Prediction for High Efficiency Vi
Beyond particle flow: Bag of trajectory graphs fordense crowd event recognition
An Efficient Surveillance Coding Method Based on a Timely and Bit-saving Background Updating Model
Low-complexity and High-efficiency Background Modeling for Surveillance Video Coding
Adaptive Moving Cast Shadow Detection
Quality Assessment for Comparing Image Enhancement Algorithms
Human Daily Action Analysis with Multi-View and Color-Depth Data
Macro-Block-Level Selective Background Difference Coding for Surveillance Video
Interactive Event Detection in Crowd Scenes
A Fast and Performance-Maintained Transcoding Method Based on Background Modeling for Surveillance V
An efficient occlusion detection method to improve object trackers
Automatic Webcam-based Human Heart Rate Measurements Using Laplacian Eigenmap
专利
检测交通事件的方法及电子警察装置
一种车标定位方法和装置
一种车辆跟踪的方法及装置
一种基于场景模型自适应更新的视频编解码方法
一种车牌图像定位的方法和装置
交通监控视频数据传输方法、装置及交通监控系统
一种基于动态贝叶斯模型网络的对象跟踪方法和设备
一种高效车辆检测方法
一种基于多视角贝叶斯网络的多视角行人检测方法
一种监控摄像机测距方法及装置
著作
<div class="page"> <div class="layoutArea"> <div class="column"> <br /> Visual Saliency Computation: A Machine Learning Perspective <
监控视频高效编码与智能分析
田永鸿的项目
2011 年国际多媒体信号处理会议
基于学习的视频关注计算及其在互动视频技术中的应用
期刊论文 11 会议论文 12 获奖 1 专利 5
　面向多摄像机联合监视的运动目标关联理论与方法研究
　面向多摄像机联合监视的运动目标关联理论与方法研究