【作者】 王娟；

【导师】 王红军；

【作者基本信息】 西南交通大学 ， 计算机技术， 2015， 硕士

【摘要】 本论文主要研究对课堂视频中举手、睡觉等异常行为的提取和对过道、楼道视频中出现的异常行为进行人脸检测和人脸识别。根据自己的研究方向,对目前工作学校的实际情况做了调查,并对系统用户(包括学校教职工、学生、学生家长等)做了详细的需求分析,确定需要实现的功能。本文设计了教室视频异常行为检测系统和过道、楼道视频异常行为检测系统的总框架和主要实现的功能模块,其均使用OpenCV2.4.4和VS2010在微软提供的MFC程序框架下实现。教室课堂视频中的异常行为检测主要用到了运动目标提取技术,论文中主要采用三帧差法实现背景分离和前景提取。在学校实际的环境情况和设备条件下,使用三帧差法检测复杂教室环境下举手、睡觉等异常目标有较好的鲁棒性。学校过道、楼道视频异常行为检测,主要是检测课间休息时间在过道、楼道中具有潜在危险性行为(如奔跑、打闹)的异常行为目标,提取出异常行为者的面部图像,进行一些图像处理操作,然后通过与我们采集的全校人脸数据库进行比对和识别,并进一步提取异常行为者的年级、班级以及姓名等信息。本系统主要调用OpenCV2.4.4中的图像处理函数进行图像处理,调用其内置的人脸检测函数进行人脸检测；然后使用HOG特征提取人脸特征,训练SVM分类器,并实现人脸识别。在实验测试中,该系统较准确的检测并标示出捕捉到的人脸,并与人脸库中的样本人脸进行识别。通过实验分析,证明本论文设计的教室视频异常检测系统和过道、楼道检测系统是稳定的,可以较好的检测出视频中的异常区域。更多还原

【Abstract】 The main research of this thesis is to extract the abnormal situation of the classroom video such as handing-up, sleep in the classroom and so on, and to detect and identify faces on the corridor video in abnormal region. According to their own research direction and the actual situation of the school, we did a detailed analysis of the system users (including teachers, students, parents of students, etc.), and determined the need to achieve the function. A detailed study of the needs analysis, I designed the classroom video anomaly detection system, corridor video anomaly detection system’s framework and the realization of function modules. The two systems are implemented by using OpenCV2.4.4 and MFC in the framework of the VS2010 provided by Microsoft.Anomaly detection in the classroom video is mainly used for moving object extraction. In this paper, three frame difference methods are adopted to realize the background separation and foreground extraction. In the actual situation of the school environment and equipment condition, the three frame difference method is used to detect the abnormal objects, such as, high performance of the complex classroom environment.School corridor video anomaly detection, is to test abnormal people who do some dangerous things during the break time and extract their facial pictures to be manipulated. Furthermore by comparison with our acquisition of university people face database. We can extract their individual information, such as class、grade and name. This system mainly uses the image processing function of OpenCV2.4.4 to carry on the image processing, and calls its built-in face detection function to carry on the face detection. And we use the HOG feature to extract the face feature, train SVM classifier, and realize the face recognition. In the experiment test, the system test and detect the identified face, to match it with data in the face database.Through the experimental analysis, classroom video anomaly detection system and corridor detection system are proved to be stable and be able to detect the abnormal regions in the video.更多还原