Internet-Oriented Video Traffic Analysis Technology

Video traffic is an extremely important part of network traffic.It contains abundant information,which is worth researching.So it will have a profound influence on practical applications to detect and identify the video traffic.As the network bandwidth is spreading widely and network traffic transmission protocols are prevailing,how to detect video streams effectively and analyze their contents accurately have already become the urgent problems to be solved.Hence,this thesis mainly focuses on the two major issues: video traffic identification and video content analysis.By and large,the main work includes the following aspects:1.Network traffic is preprocessed,namely that the disordered or the retransmitted messages are preprocessed by using the order-correcting windows.The specific procedures are as follows: First of all,a flow table needs to be set up,and then its functional organization structure should be designed as well.In the end,the order-correcting window can be operated to rearrange the disordered messages or reject the retransmitted messages in the TCP stream.Network traffic pre-processing ensures the traffic packets are in corrects orders and also lays a solid foundation for the subsequent video frame restoration.2.The transmission characteristics and message characteristics of the mainstream network video transmission protocol,including RTSP?RTP?RTMP and so on,are analyzed intensively and summarized specifically.And based on this,the identification method of the video traffic is proposed.It is verified as well via testing numerous collected traffic samples.The experimental result shows that the identification method of the video traffic,introduced in this thesis,can precisely identify over 95 percent of related video traffic and meets the requirements of the subsequent video traffic analysis.3.Based on the video single frame,a Convolutional Neural Networks model is designed to predict the UCF101 data sets.It treats video frames as images and classifies the video contents on the basis of the static characteristics of a single video frame.4.Long and Short Time Memory(LSTM)network,based on Singular Value Decomposition(SVD),is introduced to classify the video traffic contents.The spatial feature and optical-flow feature of video are extracted by Convolutional Neural Networks model,for a start;and then they are integrated into spatio-temporal feature.Ultimately the spatio-temporal feature,whose dimensionality has been reduced with the aid of the Singular Value Decomposition,is trained and predicted by inputting it to the LSTM model.Through training the LSTM model,the temporal characteristics of the spatio-temporal feature can be further extracted,making it highly efficient to classify the video contents.The network video traffic detection and video content analysis methods,proposed in this thesis,can effectively solve the problem of network video content identification,and have a certain guiding significance and practical value for practical applications.