Research On Key Technique Of Cascaded Real-time Streammedia System

Recently,in domestic market,there emerge a lot of manufacturers who focus on video surveillance industry,and their main products are video/audio compression board or card,embedded video server etc.At the same time,they have developed their own dedicated media server software platform.However,we found the following questions exist in their products:1) These systems are not compatible each other;2) Only small-scale networking is supported by these systems;3) When deployed in the Internet, for the unreliable network environment,the QoS(Quality of Service) performance of these systems is poor.Great opportunities have emerged for large-scale video surveillance system.Many industries have developed their own standards.However,the corresponding large-scale video surveillance network platforms for streaming media are not available on the market.The above problems are studied in this paper.As one of the most advanced video coding technologies,H.264 got excellent properties.We found that the intra-prediction method of H.264 does not make full use of known information.It only uses 30%～60%of availale reference pixels.In addition,it assumes that the same textures of all the pixel values are the same which does not fully reflect the texture characteristics.Therefore,we propose an improved interpolation intra prediction method.Experimental results shows that,compared with JM9.5,the proposed method increase PSNR 0.0088 dB while reduce bitrate 0.76%.In video coding,the probability of DCT coefficients being quantized to all zero is very high.If we can detect all zero macroblocks earlier,the computing resources will be greatly saved.So we propose an improved fast mode selection algorithm.Compared with JM9.5,the new method can save about 37.31%of computational time averagely and only cost the PSNR reduce about 0.054 dB.In order to deploy a large-scale internet video surveillance service network,we designed a SVSON(Surveillance Video Service Overlay Network) overlay model. SVSON has two layers SML(Service Managerment Layer) and MSL(Multimedia Service Layer).It can provide services according to customers' demands.In addition,we analyzed the QoS routing problem of SVSON and proposed a centralized load balancing heuristic algorithm,named SVSHA.SVSHA takes into account of user's functional requirements, QoS,and link resources comprehensively,and then find out the best search service links in the overlay network.Meanwhile,better load balancing can be achieved.Experimential results show that the performance of SVSHA is very close to the optimal routing algorithm.Real-time streaming media services have a strict delay requirement,while the Internet is an unreliable network.Due to the uncertainty of bandwidth,delay and jitter may occur in service time.In order to provide satisfied QoS,streaming server system must be able to detect real-time network conditions and adjust the frame rate and delivery rate from time to time.We propose a real-time feedback control scheme for streaming media QoS.It adopts TCP-Like method and dynamically adjusts the sending rate according to the current output rate and client feedback to reach a stable quality of service. Experimental results show that our algorithm can provide good QoS.To improve the efficiency of streaming multimedia caching,we realized fragmental proxy caching system.This method use perceptive file input output system which can automatically and intelligently pre-fetch subparagraph so as to ensure the continuity of media stream.According to the prevalence of media object,the system can determine the optimal allocation of cache space.Experimental results show that our algorithm provides better performance than prefix caching method.The realization of the SVSON prototype system is described in the last part of this dissertation.And the prototype had been applied to the "Guangdong Netcoffee Video surveillance system" project.And more than a thousand service nodes had been deploied in this system.During the past one year and more runtime test,this system behaved reliable and robust,and achieved satisfying performance.