Abstract: With the acceleration of urban modernization process, it’s urgent to construct an efficient and comprehensive system for managing the urban emergency. Surveillance videos obtained by monitors in multiple locations in cities are the indispensable input data of management system. However, these videos are processed by human with high costs and poor results. This paper proposes an intelligent approach for analyzing surveillance videos of urban emergency. Based on open source computer vision libraries, this approach extracts the quantities and locations of objects. Generating tracks of objects and matching pre-set abnormal templates, normalized context model of emergencies is constructed, and subsequently used for analyzing and warning. This paper also operates the proposed approach on videos acquired in districts of a university campus and high-tech park, in order to evaluate the feasibility and reasonability of the proposed approach.

Key words: urban emergency, surveillance video, intelligent analyzing, event context model

摘要： 随着城市现代化进程的不断加速，建立高效、完善的城市应急事件管理系统迫在眉睫。遍布城市各处的监控摄像头拍摄的监控视频是城市应急事件管理系统重要的数据来源，但在现阶段主要依赖人工识别，费时费力且效果不佳。提出一种面向城市应急管理的监控视频智能解析方法，利用计算机视觉库获取视频中监控对象的数量和位置，生成其运动轨迹并匹配预设的异常运动模板，从而构建应急事件的规范化情境模型用以分析和预警。以某大学校园开放区域和某高新科技园区的监控视频为例，验证了所提出方法的可行性和合理性。

关键词: 城市应急事件, 监控视频, 智能解析, 事件情境模型