无线多媒体传感器网络低复杂度视频编码及高容错传输技术研究

项目名称： 无线多媒体传感器网络低复杂度视频编码及高容错传输技术研究

项目编号： No.61471162

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 武明虎

作者单位： 湖北工业大学

项目金额： 82万元

中文摘要： 无线多媒体传感器网络中节点计算能力、存储能力和能耗受限，使得传统的预测编码框架不能满足低复杂度编码要求。易错无线环境中的压缩视频流传输也需要节点间的协作和层间协作进行高容错传输。本项目利用压缩感知理论，研究适合无线多媒体传感器网络的低复杂度分布式视频编码技术及压缩码流在有扰无线信道中的高容错传输技术。主要内容有：（1）分析视频频序列帧内结构特征和帧间远程相似性，用训练法构造视频信号的全局过完备字典，得到视频信号的非局部稀疏模型；（2）在非局部稀疏模型基础上，提出分布式压缩感知视频编解码框架，通过运动搜索和单应性矩阵补偿时间插值获得边信息，在联合解码重构中引入Huber-Markov随机场先验概率模型对解码像素点进行平滑性约束，并根据视频反馈信息进行速率控制；（3）在QoS要求下，通过节点间的协作构建分布式虚拟多天线系统，并采用跨层设计方法优化资源配置。为WMSN走向应用提供理论和技术支持。

中文关键词： 压缩感知；分布式视频编码；无线多媒体传感器网络

英文摘要： The traditonal predictive video encoding is computationally intensive, which requires significant power consumption and complexity at the sensor node. So it is not appropriate for wireless multimedia sensor networks (WMSNs). The efficient error-resilience transmission of the compressed video stream over error-prone wireless channel requires cooperation of the sensor nodes and cross-layer design scheme. The work aims to provide a low-complexity distributed compressed video sensing (DCVS) codec framework, and an error-resilience transmission scheme in lossy channels. In general, the research of this project can be summarized as follows: (1) analyzes the local and remote sparsity model of video signals, constructs the over-complete dictionary for video signals using training method, then proposes a hybrid non-local sparsity model for video signals; (2) based on the above non-local sparsity model of video signals, proposes a distributed video sensing coding scheme based on compressed sensing. The side information is obtained by motion search and compensated temporal interpolation, and the reconstructed image pixels are smoothness constraint using Huber-Markov random prior probability model; (3) constructs a distributed virtual multi-antenna system, and optimizes the allocation of resources by cross-layer design to meet the QoS (Quality of Service) requirements.

英文关键词： Compressed Sensing;Distributed Video Coding;Wireless Multimedia Sensor Networks