Research And Design Of Adaptive Transmission Of Wireless Video Monitoring System

With the development of video and audio applications which occupied muchmore network resource, especially in the wireless and mobile environment, it’s anurgent problem to guarantee the quality of video and audio so as to meet the needs ofdifferent field. In addition, it’s more challenge to solve the problem in the real-timevideo transmission like the video monitoring.Based on the video monitoring application environment, this paper concentrateson the study of adaptive transmission to ensure the quality of service. The mainmethod is to control the output of the video coding rate intelligently to fit thechanging network conditions timely. Aiming at this, this paper proposes a videocoding rate control strategy based on BP neural network, which is on the basis of theAIMD algorithm with three neural networks. Properly speaking, one neural networkis used to locate the bit rate quickly as coarse adjustment, while the other two neuralnetworks are used to fine-tune the rate, so as to obtain proper additive growth step ormultiplicative attenuation factor. According to the current packet loss rate, it needs toselect the fine-tuning mode to determine the rate.In order to realize the adaptive rate control scheme, the first should done is todetect the network environment and choose the appropriate working mode of H.264video encoder to make it achieve the goal of changing the encoder’s bit ratereal-timely. In view of the above points, this paper chooses H.264to encode thecaptured video in time by comparing the current main stream coding standard, andselect the RTP/RTCP protocol which is a streaming media transmission protocol andcan suit network well, to set up a video real-time transmission system. In this system,RTCP is mainly responsible for obtaining the situation of the current networkenvironment, and feedback to the sender packet loss rate, delay and jitter of networkinformation, as the parameters of the adaptive rate control.In addition, this paper builds a simulation network using NS2to collect theappropriate sample set for neural network training including the best target bit rateand the network performance parameters under different network conditions. In orderto make the neural network can fast convergence and accurate prediction, an appropriate data normalization algorithm is proposed, which is aimed to make thecharacteristics of dates in sample set stand out. The proposed method can reduce themean square error to1/4and make the study time reduced to5000times or so.Finally, the trained neural network is tested on-line in the video monitoringsystem. The results show that the proposed scheme can adjust the bit rate in100filmson average which is fast than the basic AIMD adaptive algorithm in ten times, whilethe basic AIMD adaptive algorithm needs thousands of films to target the desired bitrate. It presents that the proposed method performed its advantages and achieved theresearch purpose.