基于传输层的视频卡顿实时检测

发布时间：2018-12-23 09:24

【摘要】：移动互联网快速发展,特别是随着4G网络的普及,越来越多的用户观看网络视频。ISP向用户提供互联网接入服务,面对大量的互联网视频用户,若网络出现延迟、抖动或丢包等问题,网络视频容易出现卡顿现象,严重降低用户的观看体验质量。因此,ISP需要在大量视频出现卡顿前,及时改善网络条件提高用户观看体验质量。面对严峻的市场竞争,若通过终端反馈或向视频网站购买相关信息不仅困难而且成本过高。然而,ISP可以获取服务器和客户端之间的视频流量数据,如何通过网络流量改善用户观看体验逐渐受到了关注。本文主要工作是通过视频流量数据对视频播放过程进行重构并且检测视频卡顿的次数和时长,目的是帮助ISP改善用户观看体验质量。视频流量数据主要包含在HTTP和TCP数据包中。其中,HTTP数据包报文主要包含了视频的基本信息,TCP数据包报文主要包含了已编码的视频数据。然而,基于对用户隐私的保护,ISP不能解析HTTP数据包提取视频的基本信息,所以我们主要通过TCP数据包中的视频数据对视频播放过程进行重构,检测视频卡顿的卡顿次数和时长。由于TCP数据包的视频数据是无差别的已编码视频片段,我们在重构视频播放过程前,需要对视频基本信息进行估计。其中,视频清晰度和视频初始延迟的估计准确性对视频重构的影响最大,我们在文中提出了相关的估计方法。在获得视频基本信息后,我们基于队列理论提出了重构视频播放过程的数学模型和给出检测视频卡顿次数和时长的方法。实际上,网络流量数据除了视频数据之外还包括了文本、图片等数据,种类繁多而且规模庞大。ISP需要快速清洗非视频数据和检测视频卡顿信息,所以我们把我们的检测方法部署在Apache Storm实时计算系统上。我们实验的视频主要选自国内的视频网站腾讯视频、爱奇艺、优酷以及全球最大的视频网站YouTube。最后实验结果通过绝对误差、相对误差和最小二乘拟合方法评价系统的性能。结果显示我们的方法有较好的检测效果。
[Abstract]:With the rapid development of mobile Internet, especially with the popularity of 4G network, more and more users watch network video. ISP provides Internet access services to users. In the face of a large number of Internet video users, if the network is delayed, Because of jitter or packet loss, network video is prone to the phenomenon of Catton, which seriously reduces the quality of the user's viewing experience. Therefore, ISP needs to improve the network conditions and improve the quality of the user's viewing experience in time before a large number of videos appear in Carton. In the face of severe market competition, it is not only difficult but also too expensive to buy related information through terminal feedback or video website. However, ISP can obtain video traffic data between server and client, and how to improve the user's viewing experience through network traffic has been paid more and more attention. The main work of this paper is to reconstruct the video playback process through video traffic data and to detect the number and duration of video replay. The purpose of this paper is to help ISP improve the quality of user viewing experience. Video traffic data are mainly contained in HTTP and TCP packets. Among them, the HTTP packet mainly contains the basic information of the video, and the TCP packet mainly contains the encoded video data. However, because of the protection of user privacy, ISP can not parse the basic information of the HTTP packet to extract the video, so we reconstruct the video playback process mainly through the video data in the TCP packet. Check the number and duration of the video's Catton. Since the video data of TCP packets are undifferentiated encoded video fragments, we need to estimate the basic information of video before reconstructing the video playback process. The estimation accuracy of video definition and initial delay has the greatest influence on video reconfiguration. We propose a related estimation method in this paper. After obtaining the basic information of video, we propose a mathematical model of reconstructing video playback process based on queue theory and present a method to detect the number and duration of video replay. In fact, network traffic data include text, pictures and other data in addition to video data. ISP needs to quickly clean non-video data and detect video carton information. So we deploy our detection method on Apache Storm real-time computing system. The videos we experimented with are mainly selected from domestic video websites Tencent, iqiyi, Youku and YouTube., the world's largest video site. Finally, the performance of the system is evaluated by absolute error, relative error and least square fitting method. The results show that our method has better detection effect.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.8

【相似文献】