监狱犯人越界检测算法研究

发布时间：2018-06-29 03:34

本文选题：鲁棒PCA + 监狱犯人越界检测系统　；参考：《国防科学技术大学》2016年硕士论文

【摘要】：随着计算机智能技术迅速发展,安防系统向智能化方向发展开始成为可能。随着它的发展和应用,人们无需肉眼紧盯视频,避免长时间工作导致的视觉性疲劳,从而杜绝监控区域出现报警失误,防止发生违法犯罪、事故案件。此外,安防系统能充分发挥计算机视觉技术在社会舆情监控、智能交通等方面的重大作用。因而,如何运用计算机视觉技术有效、实时处理监控视频变得尤为重要,特别是监控监狱犯人。为此,针对监狱监控区域的单一背景,本文提出了一种快速鲁棒PCA方法(Fast RPCA,FaRPCA),有效学习监控区域背景,提取前景中特定行人来达到实时监控犯人的目的;另综合运用多种传统技术,还设计一套监狱犯人越界检测系统来识别狱警和犯人,避免报警失误。本文具体工作如下:(1)介绍了三种经典前景提取算法,详细分析了高斯背景建模、RPCA及Go Dec算法的基本原理和算法优缺点。(2)提出了高效的前景提取方法FaRPCA,相比RPCA和Go Dec,FaRPCA在六个基准数据集上的前景检测效率和性能更高。(3)设计了监狱犯人越界检测系统。通过集成FaRPCA前景检测算法、Canny边缘检测、霍夫直线检测、颜色识别与重心检测方法实现实时监控犯人。
[Abstract]:With the rapid development of computer intelligence technology, security system to intelligent development began to become possible. With its development and application, people do not need to focus on video with naked eyes, avoid visual fatigue caused by long working hours, so as to put an end to the occurrence of alarm errors in monitoring areas, prevent the occurrence of illegal crimes, accident cases. In addition, the security system can give full play to the computer vision technology in social public opinion monitoring, intelligent transportation and other aspects of the important role. Therefore, how to use computer vision technology effectively, real-time processing of surveillance video has become particularly important, especially for prison inmates. Therefore, aiming at the single background of prison monitoring area, a fast robust PCA method (Fast RPCA-FaRPCA) is proposed, which can effectively learn the background of the monitoring area and extract the specific pedestrian in the foreground to achieve the purpose of real-time monitoring of prisoners. In addition, a system is designed to identify prison guards and prisoners and avoid alarm errors by using a variety of traditional techniques. The main work of this paper is as follows: (1) three classical foreground extraction algorithms are introduced. The basic principles, advantages and disadvantages of Gao Si background modeling and go Dec algorithm are analyzed in detail. (2) an efficient foreground extraction method, FaRPCAA, is proposed. Compared with Gao Si and go Decn FaRPCA, the efficiency and performance of foreground detection on six datum data sets are higher. (3) The system of prison prisoner cross-border detection is designed. By integrating FaRPCA foreground detection algorithms such as Canny edge detection, Hough line detection, color recognition and center of gravity detection, real-time monitoring of prisoners is realized.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：D916.7;TP391.41

【参考文献】