无线传感网多源视频压缩感知方法研究

Due to the limitation on the power and transmission capacity of sensors, together with the dynamic change of network environment, how the video sensing application based on wireless sensor network (WSN) can meet the user experience is presently still an important and challenging issue. Geared to the characteristics and demands of the wild ecological environment sensing application based on wireless video sensor network (WVSN), this project focuses on the research on the multi-source coding model and sampling mechanism based on compressed sensing (CS), the multi-source distributed video reconstruction method, and the rate control mechanism for scalable video coding (SVC). In this project, we plan to propose a multi-source video coding model based on compressed sensing, an optimal sampling mechanism based on temporal and spatial correlation, a multi-source distributed video sparse reconstruction model, a multi-source distributed video reconstruction method, a cooperation transmission control mechanism for multi-source video stream, and a rate control mechanism for distributed scalable video coding. This project is intended to provide video monitoring service with user experience of high quality and good adaptability to network environment, sensor capability, scene variety and user demands by utilizing multi-source cooperation and cross-layer design. The above innovative research findings will provide significant theoretical basis and technical support for video applications such as ecological environment protection, salvage and disaster relief organization and so on.

由于传感节点能量和传输能力受限，加之网络环境动态变化，基于无线传感网的多源视频感知应用如何满足用户体验是一个具有挑战性的课题。本项目面向无线视频传感网实现野外生态环境感知的应用特点和需求，研究基于压缩感知的多源视频编码模型与采样机制、多源分布式视频重构方法以及可伸缩编码码率控制机制。提出基于压缩感知的多源视频编码模型、基于时空相关性的最优采样机制、多源视频分布式压缩稀疏重建模型、基于压缩感知的多源分布式视频重构方法、多源视频流的协同传输控制机制、分布式压缩视频可伸缩编码码率控制机制等。课题目标是通过多源协同和跨层设计，提供满足用户体验质量的视频监测服务，并对网络环境、节点能力、场景变化、按需服务等具有良好的自适应性。上述创新性研究成果将为野外生态环境保护、组织抢险救灾等视频应用提供良好的理论依据和技术支撑。

课题针对无线视频传感网实现野外生态环境感知的应用特点和需求，探索利用压缩感知、多源协同及跨层设计实现分布式视频编码，提升用户体验。此研究旨在突破多源分布式视频压缩方案仅适用于特定场景的局限性，提供对网络环境、节点能力、场景变化、按需服务等具有良好自适应性的视频监测服务。为此，课题采用“一体两翼”的研究思路，即首先从分布式压缩编码主体框架中两个核心——编码和解码出发，分别提出混合多重假设约束下多源视频编码方法以及基于多重假设的多源视频联合解码方法，从理论层面解决有限先验条件下的边信息估计、信息量受限情况下的视频重构等问题。其次，在上述两个分支研究基础之上，借鉴深度神经网络能挖掘大数据非线性关联性的优势，从整体设计角度入手，提出基于深度神经网络的分布式编解码新框架，解决边信息估计细节信息丢失严重、残差重构稀疏先验性无法满足、视频解码实时性差等实际难题。另外，课题还探索了分布式视频压缩感知其它关键技术，如压缩感知重构理论、帧采样率分配方案以及图像去雨雾方法等。总体而言，本项目取得了有效及较丰硕的研究成果，完成了计划任务，实现了预期目标。