基于移动网络缓存与边缘计算的VR全景视频分发技术研究

With the convergence of virtual reality (VR) and the video processing technologies, there has been an increasing number of the immersive panoramic video (PV) applications. The demand of high efficiency PV delivery technology is very urgent due to the fast progress in panoramic video content generation technologies being applied to different industries. However, delivering PV over wireless networks will be extremely challenging now due to the higher bandwidth consumption for transmitting the great volume of PV data, higher requirement of low delay response and higher requirement of computation capability of mobile terminal. Based on the traffic characteristics and content representation features of panoramic video, as well as the edge processing technologies in mobile networking, this project aims at studying the high efficiency PV delivery technologies to deal with the above-mentioned practical challenges from the following four aspects. First, we propose the high efficiency PV encoding approach based on the rate-distortion-optimized motion model selection; Second, we study the duplicate tile data placing and updating technologies in PV caching by predicting the tile popularity with fusing the visual saliency of tile, the past popularity of tile and PV popularity; Third, we study the adaptive offloading scheme of viewport projection computation based on mobile edge computing; Fourth, we propose a viewport-aware multi-user resource allocation and tile data scheduling scheme based on the coordination of radio access networking and backhaul networking. Finally, we will implement the experimental system of high efficiency panoramic video delivery over the mobile networks to verify the proposed technologies. The expected innovative research results are of the great practical significance in boosting the panoramic VR video industry.

随着虚拟现实和视频处理技术的结合，沉浸式全景视频应用逐步增多。全景视频内容制作技术与行业应用的快速结合迫切需要全景视频的高效分发技术。然而，目前全景视频在移动网络分发过程中还存在着带宽消耗大、访问延迟要求高、对移动终端计算能力要求高等实际挑战问题。面对这些挑战问题，本项目基于全景视频内容表示与应用特性以及移动网络边缘处理技术特性，从四个方面开展全景视频分发技术的创新研究：1)基于多种运动模型率失真优化的全景视频高效编码；2)基于内容显著度、分片(Tile)热度和视频流行度融合预测的全景视频空间分片数据在移动网络中的缓存副本放置与更新技术；3)基于移动边缘计算的视角投影计算自适应迁移技术；4)基于视角质量感知的蜂窝网与回程网络协同的多用户资源分配与分片数据调度技术。最终搭建全景视频移动网络分发原型实验系统，对上述技术进行验证。本项目预期的研究成果对推动虚拟现实视频产业发展具有非常现实的意义。

本项目针对移动网络下VR全景视频分发过程中带宽消耗大、访问延迟要求高、对终端计算能力要求高等挑战问题,全面分析VR全景视频移动网络分发过程中影响用户体验的关键因素，以降低通信延迟和提升视窗质量为优化目标，紧紧围绕如何优化分片缓存与视窗计算向边缘端卸载两个核心问题，聚焦信源编码/转码优化、分片缓存、边缘计算卸载、以及VR视频分片调度等研究内容，具体突破了适配移动网络状态的VR全景视频编码与网络传输联合优化、VR视频分片显著度预测与移动网络分片缓存、基于边缘计算的VR视窗渲染计算自适应卸载、视窗质量感知的多码率VR视频调度等方面的关键理论与技术，有效地提升了移动网络下VR全景视频分发的用户体验。本项目在视频处理与网络分发领域国内外权威期刊或重要会议上发表论文17篇（含IEEE/ACM Transactions期刊论文4篇），被SCI检索论文6篇，EI检索论文17篇，申请发明专利6项（3项已授权）。