| Wireless mesh networks (WMNs) are suitable for ubiquitous wide-band communicationservices because of the wide-area coverage, flexibility and low cost. It has become an importantnetworking mode of the next generation wireless communication system. However, itscharacteristics such as time-varying channel, multi-hop and node mobility will make the linksunstable, and consequently degrade the quality of service (QoS) for upper layers. Stereoscopicvideo which contains the depth information of the objects is suitable for3D natural scenes.Therefore it can provide more vivid visual experience for people. Nonetheless, we are facing thechallenges of dealing with large amounts of data and real-time transmission requirements inorder to preserve the stereoscopic video quality over mesh networks. This leads to the problem ofhow to guarantee the end-to-end video quality, and the research of this topic has importanttheoretical significance and practical application value.This dissertation mainly focuses on key technologies of end-to-end QoS for stereoscopicvideo delivery over WMNs. Taking the characteristics of stereoscopic video into consideration,several methods are proposed at different layers from the perspective of transmission networkand terminal system. We set up the model of video distortion using the channel packet lossmodel, Markov model and M/M/1/K queue model in transmission network. We select thereasonable transmission rate at Medium Access Control (MAC) layer and select the proper routeat routing layer. At the application layer of the terminal system, the decoder implements the errorconcealment for corrosion images, making use of the correlation of disparity.The contributions of the dissertation are summarized as following:(1) In order to solve the bad fish problem in a shared single channel network, we propose arate selection mechanism at MAC layer which is suitable for video transmission. First, using thecross-layer technology, MAC layer gets the requirement of frame rate and the tolerance of packetloss rate for video from application layer. Second, the maximum allowed delay of each videopacket is estimated based on the frame rate requirement, which will be used to compute themaximum number of retransmissions for a video MAC frame. Finally, within the margin of thetolerant packet loss rate of application layer, we can adaptively select the highest possibletransmission rate for video MAC frames in terms of current channel quality and the maximumnumber of retransmissions. The lost packet will be dealt with using the algorithm of errorconcealment in the decoder. Experimental results show that, compared with the algorithm ofchanging contention window, the proposed method can reduce the delay and jitter of video service when the transmission rate should change during the Signal to Noise Ratio (SNR) ofvideo node declines, and make the throughputs of the transmission hosts degrade gracefully.Therefore, it increases the quality of reconstructed video to a certain extent and eliminates theperformance anomaly of network effectively.(2) Aiming at the problem of several routes between the source node and the destinationnode, we use video distortion as the metric to predict the transmission distortion of a route, andselect the route with minimal distortion as the transmission route in terms of the functionalrelationship between the distortion and the network packet loss and delay. The proposedalgorithm considers the factors of channel error, route interference and network congestion. First,random uniform model and Gilbert-Elliott (GE) model are used to simulate the random and burstpacket loss of the channel. Second, Markov model is used to calculate the packet collisionprobability of MAC layer and the processing time, and M/M/1/K queue system is used to predictthe delay and the probability of packet loss. Finally, we calculate the distortion of the total routeand select the route with minimum distortion of the network. Experimental results show thatcompared with the Dynamic Source Routing (DSR) protocol, our algorithm can provide betterobjective and subjective quality of video.(3) In order to eliminate the random errors for stereoscopic images in video applicationlayer of terminal system, we use the disparity correlation between pixels to propose an errorconcealment algorithm based on local reliable disparities. This algorithm uses the pixelssurrounding the lost block. First, in order to get accurate disparities, we design abase-point-biased window, and use different windows in terms of the position of estimated pixelto perform the adaptive-weight disparity matching. Second, reliable disparities of local area arefigured out according to the principles of disparity constancy of left-right consistency. And thenthe winner-takes-all strategy is adopted to remove the error disparities. These two strategiesmake the estimated disparity of the lost block better. Finally we use the estimated disparity torecover the lost block. Experimental results show that compared with other algorithms, theproposed method can significantly improve the Peak Signal to Noise Ratio (PSNR) value as wellas subjective quality with almost the same computational complexity.(4) In order to solve the random and consecutive errors for stereoscopic images, we use thedisparity correlation of blocks to propose an error concealment algorithm based on thesmoothness of boundary. First, we select the up, down, left and right blocks around the lost block,and conduct the disparity matching for them. Second, we use these four disparities to conceal thelost block, and calculate the smoothness of the boundary respectively. Then we select the bestsmoothness as the probable concealment result. If the best smoothness is still beyond the threshold, we carry out the content adaptive judgment for the lost block. The lost block will beclassified into one of the three types which are the texture block, directional block, and occlusionblock. We use different algorithms to conceal the lost block according to its type. Experimentalresults show that the proposed method substantially outperforms other algorithms both visuallyand in terms of PSNR, and at the same time, it has a lower computational complexity. |
PDF Full Text Request