面向城域视频的多目标人车混合行为理解关键技术研究

项目名称： 面向城域视频的多目标人车混合行为理解关键技术研究

项目编号： No.61472019

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 盛浩

作者单位： 北京航空航天大学

项目金额： 80万元

中文摘要： 针对传统分级分区域的视频监控系统普遍存在着自动化处理能力薄弱、跨摄像机内容关联不畅、大范围视频内容分析缺乏统一时空模型、异常行为感知手段匮乏等问题，本课题开展基于统一时空模型的多目标检测与跟踪、跨摄像机接力跟踪、异常事件感知的探索研究。其主要研究内容如下：①基于城域地理信息的视频统一坐标描述与映射；②基于动态社会力模型的人车混合跟踪；③基于交通仿真约束的跨摄像机时空建模；④基于区域统计集合的目标异常行为检测方法。通过关键技术探索和创新，建立城域视频的统一坐标描述与映射，实现人车混合跟踪和跨摄像机多目标跟踪，感知异常行为。最后，实现面向城域视频的多目标人车混合行为理解原型系统，验证上述方法和技术的有效性和实用性。

中文关键词： 视频投影；多目标跟踪；跨视域跟踪；行为感知；视频分析

英文摘要： In view of the traditional hierarchical and regional video monitoring system, the ability of automatic processing is not satisified. The content relation across the camera is not smooth, as such the unified time-space model lack capability for the analysis of large-scale video content. The perception for abnormal behavior is not capable. Thus, based on the unified time-space model, the research aims at the multi-target detection and tracking, relays tracking across the camera, perception of abnormal event. The main research contents are as follows: (1) description and mapping of video coordinate system based on urban traffic information; (2) pedestrian-vehicle mixing tracking based on dynamic social force mode; (3) time-space modeling across the camera based on traffic simulation; (4) abnormal behavior detection based on Local Statistical Aggregates. Through the exploration and innovation of the above key technologies, unified video coordinate description and mapping for urban traffic is eastablished, and pedestrian-vehicle mixing tracking and multi-objective tracking across the camera can be realized to detect the the abnormal behavior. Finally, this research constructs a prototype system of multi-objective pedestrian-vehicle mixing behavior understanding for verifying the validity and practicability of the above technologies.

英文关键词： Video projection;Multi-target tracking;Cross-camera tracking;Visual perception;Video analysis