基于分布式压缩感知的无线视觉传感器网络多视点信息有效获取相关机制研究

Wireless Visual Sensor Networks (WVSN) can provide content-rich multi-view visual information and would have broad potential applications in the future intelligent Internet of Things (IOT). The traditional multi-view information acquisition architecture with in-network processing and layered transmission severely limits the practical application of energy- and bandwidth-limited WVSN. This project aims to develop a new architecture and related methods of multi-view information collection, transmission and reconstruction for WVSN based on the emerging Distributed Compressive Sensing (DCS) technique. The research content includes three aspects. To avoid communication overhead in in-network node management and rate control, visual nodes clustering based on compression-region manifold-learning in the receiver and measurement rate control based on the state of the visual queue of the receiver are firstly studied; optimal distributed cross-layer transmission and its global convergence are realized through dual decomposition and reverse back-pressure techniques; new methods for joint design of multi-view sensing matrix and redundant dictionary and joint reconstruction of compressive sensed multi-view information based geometric transformation are proposed to reduce sensing rate and farther optimize transmission performance. This project would develop a new architecture with features of 'distributed light-weight in-network compression, cooperative and optimized transmission and joint processing at backend receiver' and a suit of creative methods for multi-view visual information acquisition in WVSNs, which is expected to significantly increase useful lifetime of WVSN and accelerate its application in the intelligent IOT, and have significant scientific and practical value.

无线视觉传感网络能够提供内容丰富的多角度视觉信息，在未来智能物联网中有着广泛的潜在应用。传统的"在网处理，分层传输"的信息获取框架制约着能量和带宽受限的无线视觉传感网络的实用性。本项目拟把分布式压缩感知应用到无线视觉传感网络中，研究多视点信息获取的新架构和新方法。研究内容包括：为避免在网节点管理和速率控制的交互开销，研究基于接收端压缩域流形学习的视点分簇和基于接收端虚拟队列的感知速率控制策略；通过对偶分解和反向背压机制实现多视点压缩感知信息的分布式跨层优化传输及其全局收敛性；通过多视点感知矩阵和冗余字典的联合设计以及基于相关几何变换的多视点联合重建降低感知速率要求，优化传输性能。本研究能够实现一种"分布式轻量在网压缩，协同优化传输，后端联合处理"的多视点信息获取新架构及一系列创新性的相关方法，从而提高无线视觉传感网络的有效运行时间，促进其在智能物联网中的应用，具有重要的科学意义和实用价值。

本课题围绕多视点无线视觉传感器网络的信息有效获取问题，完成了在视觉节点压缩感知速率控制、不同网络环境下的分布式跨层传输优化以及多视点压缩感知的联合重建等三个层面的研究工作。具体研究成果包括：根据多视点无线视觉传感网络的特点，提出接收端基于虚拟队列的压缩感知速率控制策略，有利于降低系统控制开销；在无线视觉传感网络分布式传输优化方面，提出了一种无线视觉网络中可伸缩编码视频的分布式传输算法并验证了其收敛性和最优性。提出一种基于认知无线电的压缩感知无线视觉网络中感知质量和网络传输时延的权衡优化模型以及基于随机优化理论的分布式机会传输算法，该算法不需要预知可用频段的概率分布。提出一种基于中继的OFDMA蜂窝网视觉信息传输的跨层优化框架以及实现了一种半分布式的算法，该算法通过两层对偶分解大大降低资源分配的计算复杂度。通过引入虚拟能量队列模型，提出基于能量收集的压缩感知无线视觉传感网络动态资源分配算法并提供了该算法的稳定收敛性和渐进最优性理论。在多视点分布式压缩感知联合重建及其优化方法方面，初步利用接收端多次循环重建加运动估计的方法进行多视点分布式压缩感知的联合重建，并提出面向传输优化的基于接收端虚拟队列速率控制与联合重建的融合方案。