面向无线多媒体传感器网络的高效压缩视频感知

Distributed compressed video sensing (DCVS) is a framework that integrates both compressed sensing and distributed video coding to achieve a low-complexity video coding, and the encode data is fault-tolerant and scalable which make it ideal for wireless multimedia sensor networks (WMSNs) applications. However, there is still a large performance gap between the DCVS and the traditional video encodec. Block and frame-based strategy used in distributed compressed video sensing which separates the internal relations in data is an important cause of performance loss. In response to this situation, returning of unity in reconstruction of the video and the corresponding methods of measurement are proposed. Based on tensor decomposition, as the study entry point, and the intrinsic characteristics of sparse-multidimensional in the node video and multi-view video, the one-dimensional compressive sensing theory is extended to multi-dimensional through sparse tensor decomposition.Then the measurement strategy of multi-dimensional compressed sensing will be explored . And combining dictionary learning, structured sparse and some other CS recovery theories, reconstruction algorithms of multidimensional video under constraint of sparse tensor will be studied as the main topic of this project to try and improve the quality of single-view video, multi-view video in WMSNs. The project will not only provide a new paradigm of video encodec of low-complexity, high-quality reconstruction,but also a new exploration of multi-dimensional compressed sensing theory.

结合压缩感知与分布式编码的分布式压缩视频感知由于其编码简单，编码数据具有容错好、可裁剪等特点，使得其非常适合无线多媒体传感网（WMSNs）应用。但现有采用分块压缩感知的分布式压缩视频感知与传统编解码方法在性能上仍有较大差距，分块分帧的方式割裂了数据内在联系，是造成性能不足的重要原因。针对这一情况，提出视频回归整体的重构及相应测量方法。课题以张量分解为切入点，根据节点视频及多视点视频内在的空时稀疏、多维的特点，将一维压缩感知理论通过稀疏张量分解扩展到多维，探索多维压缩感知的测量方法，结合字典学习、结构化稀疏等理论重点研究稀疏张量约束下的多维视频重构算法，从而尝试提高WMSNs系统的单节点视频、多视点视频的恢复质量。项目将为WMSNs提供一个低编码复杂度、高恢复质量的视频编解码新方案，同时也是多维压缩感知理论的新探索。

无线多媒体传感器网络是在传统无线传感器网络(WSNs)基础上引入了音频、视频、图像等多媒体信息感知功能的一种新型传感器网络，其是一种能耗敏感的无基础设施网。由于WMSNs硬件节点计算能力、能量等方面受到的限制，在传输多媒体数据特别是视频数据之前必须对其进行压缩编码，减少数据传输量以节约能量。如何同时保持较低的能量消耗及可靠的图像质量，需要在计算复杂度和通信数据量之间达到平衡，这是亟待解决的问题。. 结合压缩感知与分布式编码的分布式压缩视频感知由于其编码简单，编码数据具有容错好、可裁剪等特点，使得其非常适合无线多媒体传感网（WMSNs）应用。但现有采用分块压缩感知的分布式压缩视频感知与传统编解码方法在性能上仍有较大差距。针对这一情况，课题以张量分解为切入点，根据节点视频及多视点视频内在的空时稀疏、多维的特点，将一维压缩感知理论通过稀疏张量分解扩展到多维，探索多维压缩感知的测量方法，研究稀疏张量约束下的多维视频重构算法。项目的主要研究内容主要有以下几点，首先研究了基于克罗内克积的分块压缩感知方法，此方法可以将多维信号纳入统一的压缩感知模型，同时具有较低的复杂度、较少的感知矩阵存储量。其次是重点研究了图像、视频的重构算法，项目提出了四种新的具有稀疏、低秩张量约束的压缩感知重构算法模型。这四种新算法都相比传统的压缩感知重构算法在峰值信噪比指标上具有较大提升。最后项目研究了率失真模型，进行初步的率失真曲线拟合，并利用树莓派作为多节点，初步验证了压缩感知重构的实现性能。. 项目从应用层的视频编解码层面探讨了基于压缩感知理论的新型的重构算法，并进行了验证，相关理论可以为图像逆问题提供相关参考，其也是多维压缩感知理论的新探索。同时项目研究涉及编解码的整个流程，可为WMSNs提供了一个低编码复杂度、高恢复质量的视频编解码新方案的参考。