基于异常事件检测的交通监控视频摘要

发布时间：2018-02-01 03:51

本文关键词： 目标检测目标跟踪梯度方向直方图异常事件检测随机森林视频摘要　出处：《浙江大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着交通场景中监控摄像头的数目越来越多,监控视频数据迅速增长,然而监控场景中发生的事件只有一部分才是人们关注的重点。本文将人们关注的一类事件定义为异常事件,并提出算法将视频中的异常事件整合成视频摘要概要地展示视频的人们关注的主要内容。本文提出基于异常事件的交通视频监控方法主要分成三个部分,分别是检测跟踪得到运动事件、通过分类检测到异常事件以及整合成视频摘要。对给定视频中的视频帧提取特征结合带隐变量的支持向量机逐帧检测视频中的目标,匹配检测结果得到视频中运动轨迹,将其表示为运动事件。分析其中异常事件的特点建立一个结合Bootstrapping的随机森林模型做分类。最后提出基于关键帧以及基于关键区域的视频摘要提取算法。基于关键帧的视频摘要提取算法通过筛选包含异常事件检测区域的视频帧,然后整合成新视频实现。基于关键区域的提取算法实现了目标关键区域的最优化重组,拼接组合成新的图像帧生成视频。本文中提出的方法针对交通场景中的异常事件进行分析,最终实现视频摘要的算法。实验表明该算法具有较高的检测准确率及计算效率,实现了视频的精简压缩。
[Abstract]:With the increasing number of surveillance cameras in traffic scenes, video surveillance data is growing rapidly. However, only a part of the events in the monitoring scene are the focus of attention. In this paper, we define a class of events as abnormal events. An algorithm is proposed to integrate the abnormal events in the video into the main contents of people's attention in the video summary. In this paper, the traffic video surveillance method based on the abnormal events is mainly divided into three parts. The tracking is used to detect and track the motion events. Abnormal events are detected by classification and integrated into video abstracts. Feature extraction of video frames in a given video is combined with support vector machine with hidden variables to detect targets in video frames. The motion trajectory in video is obtained by matching detection results. By analyzing the characteristics of abnormal events, a stochastic forest model combined with Bootstrapping is established for classification. Finally, a view based on key frames and critical regions is proposed. Frequency digest extraction algorithm. Video digest extraction algorithm based on key frames filters video frames containing abnormal event detection region. Then it is integrated into a new video implementation. The extraction algorithm based on the key region realizes the optimal recombination of the target key region. The method proposed in this paper is based on the analysis of abnormal events in traffic scene. The experiment shows that the algorithm has high detection accuracy and computational efficiency and realizes the reduction of video compression.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【相似文献】