交通路口智能视频监控系统设计

发布时间：2018-05-02 08:53

本文选题：改进的高斯背景模型 + 阴影检测　；参考：《华中科技大学》2014年硕士论文

【摘要】：进入21世纪，城市公共安全已经越来越成为人们所关注的问题。这一切，都迫切要求发展城市安防监控系统。传统的城市安防监控系统采用的是人工监控的方式，这种监控方式随着城市监控摄像头的个数的激增存在着工作量过大的问题。因而，安防监控系统的智能化迫在眉睫，本文的目标即是实现了一个这样智能的、无人工干预的、基于视频的城市安防监控系统。系统主要实现的功能包括实时在线的异常监控与事后取证时智能视频检索。系统的实现方案主要包括运动目标检测、运动目标跟踪、异常检测三个核心模块。在运动目标检测模块，，本文采用的方法是改进的混合高斯背景建模。针对传统的混合高斯背景建模存在的第一帧初始化不全是背景会出现“鬼影”、原始方差的更新策略会导致噪声过多、算法计算存在冗余这三个问题，本文分别相应提出了更新率自适应、方差设定阈值以及K值自适应的改进策略。对于运动检测中阴影、光线的矫正以及后处理模块，本文也都相应提出了解决方案。在运动目标跟踪模块，本文基于经典的团跟踪进行改进，针对经典团跟踪算法中团的对应方式中缺失的团的多对多对应关系情形本文予以补充，而针对团跟踪在静态遮挡和动态遮挡情形下失效的情况，本文提出了相应的线性预测和分块跟踪的解决方案。在异常检测模块，本文通过事先的异常定义结合目标的位置、轨迹与运动状态从而能够实时在线的检测出异常。对于事先不能够预测的异常，本文提出予以事后视频检索的方式进行补充。最后，通过系统测试与功能测试模块证明，本文提出的智能视频监控系统能够实现预先设定的功能，并且有较好的实时性与优越的算法性能。
[Abstract]:In the twenty-first Century, urban public security has become more and more concern. All of these are urgently required to develop the urban security monitoring system. The traditional urban security monitoring system is used by artificial monitoring. With the increase of the number of urban surveillance cameras, there is a large amount of work. Therefore, the intelligence of security monitoring and control system is imminent. The aim of this paper is to realize an intelligent, non intervention, video based urban security monitoring system. The main functions of the system include real-time online anomaly monitoring and intelligent video retrieval when taking evidence afterwards. The realization scheme of the system mainly includes movement. Target detection, moving target tracking, anomaly detection three core modules. In the moving target detection module, the method used in this paper is an improved hybrid Gauss background modeling. For the traditional mixed Gauss background modeling, the first frame initialization is not complete in the background, the background will appear "ghost shadow", the original variance updating strategy will lead to the noise. There are three problems of redundancy in algorithm computing. In this paper, the adaptive updating rate adaptive, variance setting threshold and K value adaptive strategy are put forward respectively. In motion detection, the shadow, light correction and post-processing module are also proposed. In the motion target tracking module, this paper is based on the classic. The regiment tracking is improved to supplement the multiple to multi corresponding relationship of the missing groups in the regiment's correspondence method in the classical regiment tracking algorithm, and for the case of the failure of the regiment tracking in the static occlusion and dynamic occlusion, the corresponding linear prediction and block tracking solution is proposed in this paper. In this paper, the abnormity can be detected in real time by combining with the location of the target, the trajectory and the state of motion in advance. For the exception that can not be predicted in advance, this paper puts forward a method to supplement the video retrieval after the event. Finally, it is proved by the system test and function test module that the intelligent video surveillance proposed in this paper is proposed. The system can achieve pre set function, and has better real-time performance and superior algorithm performance.

【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN948.6

【参考文献】