复杂运动场景视频大数据中异常事件检测研究

项目名称： 复杂运动场景视频大数据中异常事件检测研究

项目编号： No.61502042

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 梁美玉

作者单位： 北京邮电大学

项目金额： 20万元

中文摘要： 利用计算机视觉和视频智能处理技术对旅游景区视频进行实时监控和异常事件检测，构建智能化的旅游突发事件监测系统，对于保障旅游安全具有重要意义。本项目在以往科学研究基础上，提出基于时空显著性感知的变换域跨尺度自适应视频增强方法；构建能够适应于复杂运动场景的快速鲁棒性时空非局部模糊配准机制，提出基于关联学习和时空非局部相似性的视频超分辨率重建方法；针对视频大数据中的背景混杂、遮挡、噪声干扰、视角变化、光照变化等复杂场景，提出能够适应于复杂场景下的鲁棒性多目标跟踪方法。通过提取目标区域的鲁棒时空特征，并结合概率主题模型，构建目标行为的鲁棒性时空语义特征空间，建立复杂运动场景视频大数据中的异常事件检测模型；构建面向视频大数据的旅游突发事件监测系统，为及时进行旅游突发事件预测和预警、保障旅游安全提供有力的技术支撑，力争在复杂运动场景视频大数据中异常事件检测领域取得突破性进展。

中文关键词： 视频大数据；时空显著性；超分辨率；异常事件检测；旅游突发事件

英文摘要： It has great significance for the protection of tourism safety that we must make real-time monitoring and detection of abnormal events for the tourist attractions video and construct intelligent tourism emergency monitoring system by using computer vision and intelligent video processing technologies. Based on the past research experience, this project will propose transform domain trans-scale adaptive video enhancement method based on spatio-temporal saliency sensing; construct fast and robust spatio-temporal nonlocal fuzzy registration mechanism which can be adaptive to complex motion scenes, and propose video super-resolution reconstruction method based on correlation learning and spatio-temporal nonlocal similarity; for the complex scenes in the video big data such as mixed background, occlusion, noise disturbance, perspective changes, illumination changes, propose robust multi-target tracking method which can be adaptive to complex scenes; by extracting robust object region spatio-temporal feature and combining the probability topic model, construct robust spatio-temporal semantic feature space of object behavior, and establish abnormal event detection model for video big data under complex motion scenes; construct tourism emergency monitoring system for video big data. This project will provide strong technical support for the timely tourism emergency prediction and tourism safety protection, and strive to achieve breakthrough progress in abnormal event detection for video big data under complex motion scenes.

英文关键词： video big data;spatio-temporal saliency;super-resolution;abnormal event detection;tourism emergency