多跳网络中基于内容重要性的视频传输资源分配技术

Content-aware resource allocation has recently emerged as a very active research topic in area of video transmission over multi-hop networks. Facing the technical challenges therein, this project studies the bandwidth allocation and packet scheduling strategies for video transmission over multi-hop networks while taking into consideration the impact of video contents on the video quality. The novelty and technical contributions are as follows: (1) to address the problem of accurate measurement the content importance of each video packet on the video quality, we study the propagation behavior of the transmission distortion and aim to find the intrinsic relations of the distortion and the coding parameters. We attempt to propose a 3D motion reference map to quantify the distortion propagation behavior and then a low complexity yet accurate transmission distortion model to predict at the encoder the potential transmission distortion of each video packet once it is lost; (2) we study the problem of effective packet scheduling for video transmission over multi-hop networks, in which important impact factors, such as the different contributions of video packets to the reconstructed video quality, stringent delay constraints and dynamic network conditions, are to be comprehensively considered; (3) we analyze the impact of the different input video streams，dynamic network conditions and bandwidth allocation as well as stringent delay requirements of different types of video applications on the packet loss probability for video streaming over multi-hop networks and address the problem of end-to-end packet loss probability prediction for video streaming over multi-hop networks using machine learning-based methods; (4) we study the problem of bandwidth allocation for multi-session video transmission over multi-hop mesh networks, where multiple video streams share a network and compete for the limited bandwidth resources, with the aim to minimize the overall distortion of multiple videos sessions while balancing the video qualities among them. To solve this problem, we firstly analyze the impact of different bandwidth allocation on the reconstructed video quality and develop a resource-distortion model for video streaming over multi-hop networks with content-aware packet scheduling scheme. We then attempt to formulate the bandwidth allocation problem as a constrained multi-objective optimization problem and will try to use evolutionary algorithms to find the optimal resource allocation. The proposed algorithms can be can be further applied to various video transmission applications over multi-hop networks to enhance the utilization of the limited network resources, which is of great importance, both academically and industrially, for the efficient multi-user video transmission over resource-limited multi-hop networks.

基于视频内容的网络资源分配是近年来视频传输领域的研究热点。本项目旨在视频内容重要性准确度量的基础上，研究多跳网络中资源受限时基于内容重要性的视频传输带宽分配与数据包调度问题。针对其中的关键问题和技术难题，研究并提出有效的解决方案：(1)针对现有模型在失真扩散特性本质分析和定量度量方面的欠缺，提出基于三维运动参考图的包传输失真模型，实现数据包内容重要性的准确预测；(2)提出视频内容和时延约束相结合的包调度策略，兼顾视频应用的解码质量和实时性需求；(3)针对队列理论在多跳网络丢包率分析方面的欠缺，提出基于学习的多跳网络丢包率模型，实现复杂多跳网络丢包率的有效预测；(4)提出基于资源-失真优化的多跳网络带宽分配技术框架和算法，兼顾多视频总体质量和公平性的同时，实现有限带宽的合理分配和高效利用。研究成果可应用于各种多跳网络视频传输应用，对资源受限条件下的高效视频传输具有重要的理论意义和实用价值。

本项目在视频内容分析、内容重要性建模与预测的基础上，系统地研究了多跳网络中资源受限时基于内容重要性的视频编码码率控制与分配和传输带宽资源优化分配等关键问题，提出了一系列有效的解决方案：(1)提出了一种基于三维运动参考图的传输失真描述与预测模型，实现了宏块级、数据包级和帧级等多粒度视频内容重要性的准确预测；(2)提出了一种视频内容和时延约束相结合的包调度策略和一种基于视频内容重要性的不等丢包保护策略，能合理有效地分配有限的编码码率和传输带宽资源，实现了网络视频系统的性能优化；(3)提出了一种基于学习的多跳网络丢包率预测模型，实现了在基于视频内容和时延约束相结合的包调度策略情况下、视频码流在复杂多跳网络中传输时丢包率的有效预测；(4)提出了一套基于内容的网络视频传输系统带宽资源分配和性能优化决策模型和求解算法，实现了多个视频流之间以及每个视频流内的带宽资源分配和数据包优化调度，提高了有限带宽和严格时延约束情况下的网络视频传输系统性能；(5)提出了一种适应大范围带宽变化的灵活编码控制策略和基于内容重要性的帧级和CTU级质量平稳码率控制算法，实现了大范围带宽自适应的灵活和质量平稳码率控制；(6)针对最新的高效视频编码标准HEVC复杂度高难以推广应用的难题，提出了一套基于视频内容的低复杂度实时帧间帧内预测、变换量化和环路滤波优化算法，对下一代视频编码技术的发展及其应用具有重要理论意义与应用价值。.本项目部分成果已用于某型号边海防视频传输与处理系统改进，该系统已在全军边海防哨所安装300余套；部分成果已用于某新型号边海防视频传输与处理系统（已完成设计定型），有望在不久的将来在我国边海防地区应用，预计全军装备400余套，提升视频情报在实际边海防复杂网络中的可靠稳定传输，为视频情报的智能分析研判等后续处理奠定基础，社会、经济或军事效益显著。