情绪视频监控系统中的人脸采集服务器设计与实现

发布时间：2017-12-28 01:02

本文关键词：情绪视频监控系统中的人脸采集服务器设计与实现　出处：《华南理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着社会的发展,工作、生活的节奏越来越快,现代人面临的压力也越来越大。个人由此产生的心理问题也是愈来愈多。在这种现实情况下,在一些特殊场合,比如,学校、养老院等场所需要通过识别学生、老年人的情绪来了解其心理状态从而为其提供心理引导的需求变得越来越迫切。因此需要一种通过视频来识别被拍摄者情绪的技术来满足这种需求。情绪视频监控系统是一种对从视频中采集到的图片使用机器学习相关算法来分析被拍摄者的情绪的新型应用。情绪视频监控系统的服务端包括人脸采集服务器、流媒体服务器、算法服务器、Web服务器和数据库服务器,其中人脸采集服务器负责从实时视频流中采集图片发送至算法服务器。因为机器学习算法识别人脸情绪需要一定耗时,所以人脸采集服务器选取单线程采集、多线程发送,采集发送线程之间共享内存队列的架构,使之在相关机器学习算法识别人脸情绪需要一定耗时的情况下,系统仍然能够保证吞吐量维持在一个较高的水平。用户在终端界面观看人脸情绪识别结果的同时,也能够观看视频源的实时视频流。由于视频源(网络摄像头)只支持有限路数的输出,所以在观看视频流的用户增多时,为了使得人脸采集服务器采集到的人脸图片质量不会下降(人脸图片质量下降会影响人脸情绪分析结果),用户应该观看由流媒体服务器转发的视频源的实时视频流。人脸采集服务器和流媒体服务器均能够在运行过程中接收来自Web服务器的请求并作出相应处理,同时作为运行在真实环境下的服务端程序,人脸采集服务器和流媒体服务器均能自动从异常情况下恢复并且重新读取已有数据提供服务。如上所述,本文实现了一个在算法分析人脸情绪需要一定耗时的情况下仍然保持高吞吐量的人脸采集服务器和一个实时转发直播视频流的流媒体服务器。测试结果表明,人脸采集服务器和流媒体服务器功能和性能指标均达到预期水平。
[Abstract]:With the development of society, the pace of work and life is getting faster and faster, and the pressure of modern people is becoming more and more serious. There are more and more psychological problems arising from the individual. In such a real situation, in some special occasions, such as schools, nursing homes and other places, it is urgent to understand their mental state and provide psychological guidance for them by identifying the emotions of students and seniors. So there is a need for a video to identify the moods of the photographed person to meet this requirement. Emotional video surveillance system is a new type of application that uses machine learning algorithms to analyze the emotions of the photographers from the pictures collected from the video. The server of emotional video surveillance system includes face collection server, streaming media server, algorithm server, Web server and database server. The face collection server is responsible for collecting pictures from real-time video stream and sending it to algorithm server. Because the machine learning algorithm for face recognition requires a certain emotional time, so face acquisition server selection single thread acquisition, multi thread transmission, memory queue architecture is shared between collecting and sending thread, the learning algorithm of face recognition requires a certain emotion time-consuming in the relevant machine, the system can still guarantee the throughput is maintained at a high level. At the end of the user interface, the user can watch the result of the face emotion recognition, and can also watch the real-time video stream of the video source. As the video source (webcam) output support only a limited number, so watch streaming video users increased, in order to make the face image quality will not face acquisition server collected decreased (face image quality decline will affect the results of sentiment analysis, users face) should be viewed by real-time video streaming media server to transmit the video source flow. Face acquisition server and streaming media server can receive requests from the Web server in the operation process and make the corresponding treatment, at the same time as the operation in the real environment of the server program, image acquisition server and streaming media server can automatically from the abnormal cases recovered and re read the existing data to provide services. As mentioned above, this paper implements a face collection server that maintains high throughput and a streaming media server that transmits live video streaming in real time under the condition of algorithm analysis. The test results show that the function and performance indexes of the face acquisition server and the streaming media server all reach the expected level.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN948.6

【相似文献】