视频监控中实时人脸识别系统的研究与实现

发布时间：2018-01-16 06:22

本文关键词：视频监控中实时人脸识别系统的研究与实现　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

【摘要】：高考是国家选拔人才的重要手段。为了保障考试公平公正,在高考录取工作中,只能允许指定人员进入工作场地。为了解决录取入场人员身份识别问题,本课题提出了对录取工作场地入口采用视频监控,在计算机捕获的视频图像中对出现的人脸进行身份识别。为了防止伪造的人脸图像欺骗机器通过自动人脸识别,视频人脸识别中必须再加以活体检测技术判断视频中的人脸是否是真实的人脸。笔者通过研究相关文献,对视频中面部基准点变化的分析。提出了基于归一化眼眶距离比描述人眼眶距离在视频中的变化。通过计算其方差与一定的阈值相比较,可以对视频中的眨眼行为进行监测,进而确定是否是真实的人脸。该活体检测方法计算容易,占用内存少,而且不需要添加额外传感器,在测试中得到92.4%以上的准确率。实验分别从效率和精度两方面对人脸检测校准和人脸识别方法进行甄选,最终选择了基于形状索引特征的联合层叠模型的人脸检测校准方法和基于深度学习的GoogLeNet人脸识别方法。在VS2013软件平台上使用.NET、EmguCV、FaceSDK等类库编写完成了一个准确高效的视频监控实时多人脸识别系统。该系统满足了对高考录取场地人员身份验证的需求,于2016年河南省高考录取工作中进行了部署,取得了良好的实际应用效果。
[Abstract]:The college entrance examination is an important means for the national selection of talents. In order to ensure the fairness and fairness of the examination, only the designated personnel can be allowed to enter the work site in the work of the college entrance examination, and in order to solve the problem of identity identification of the admitted persons. In this paper, video surveillance is proposed for the entrance of the admission work site. In order to prevent the fake face image spoofing machine from automatic face recognition. In video face recognition, it is necessary to use live detection technology to judge whether the face in the video is a real face. The change of facial reference point in video is analyzed. Based on the normalized orbital distance ratio, the variation of human orbital distance in video is described. The variance is compared with a certain threshold by calculating the variance. The blinking behavior in video can be monitored to determine whether the face is real. This method is easy to calculate, takes up less memory, and does not need to add additional sensors. In the test, the accuracy is more than 92.4%. The method of face detection and calibration and face recognition are selected from the efficiency and accuracy of the experiment. Finally, the method of face detection and calibration based on joint stacking model based on shape index feature and the method of GoogLeNet face recognition based on depth learning are selected. The use of .NE on VS2013 software platform. T. An accurate and efficient video surveillance real-time multi-face recognition system has been developed by using the class libraries such as the EmguCVV Facebook SDK. The system meets the needs of the authentication of the personnel in the entrance examination venue. In 2016, Henan Province College entrance examination admission work was deployed, and achieved good practical application results.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN948.6

【相似文献】