SBIR Phase I: Domain and Network Aware High Definition (HD) Videoconferencing

这项小企业创新研究（SBIR）第一阶段提案旨在为高清（HD）视频会议创建一种低带宽流算法。最终目标是一个软件系统，实现小于150毫秒的单向端到端延迟（电话的典型延迟）1280x720 @ 30 fps视频在512 Kbps（高端消费者宽带的典型上传速度）。如今，大多数视频会议系统的视频流分辨率都是高清的1/10。在这样的分辨率下，通常很难看到面部表情。商业高清产品目前只在硬件上可用。这些系统都需要超过宽带限制的网络速度。本课题旨在开发宽带视频会议软件。面孔和白板是视频会议中最常见的两个感兴趣的元素。这项研究将结合实时网络感知和人们如何感知面部表情、白板和非描述性背景的领域知识来创建一个独特的压缩/流算法。输入视频被分割成不同的区域。白板和背景将使用新颖的超分辨率技术增强清晰度。将开发两项核心技术：1)增强白板和背景的感知分辨率；2)网络自适应高清流媒体。尽管在过去的几十年里，刚刚描述的概念在原则上已经被成千上万的研究所探索，但之前的努力并没有产生任何HD产品。

This Small Business Innovation Research (SBIR) Phase I proposal aims to create a low bandwidth streaming algorithm for high definition (HD) videoconferencing. The ultimate goal is a software system that achieves less than 150 msec one-way end-to-end delay (the typical delays of telephone) for 1280x720 @ 30 fps video at 512 Kbps (the typical upload speeds of high-end consumer broadband). Most videoconferencing systems today stream video at 1/10 of HD resolution. At such resolution, it is often difficult to see facial expressions.Commercial HD products are currently only available in hardware. These systems all require network speeds beyond the limits of broadband. This research aims to develop videoconferencing software for broadband. Faces and whiteboards are the two most common elements of interest in a videoconference. This research will combine real-time network sensing and the domain knowledge of how people perceive facial expressions, whiteboards, and nondescript background to create a unique compression/streaming algorithm. The input video is segmented into different regions. The whiteboard and background will be enhanced for sharpness using novel super resolution techniques. Two core technologies will be developed: 1) perceptual resolution enhancement of whiteboard and background, and 2) network adaptive HD streaming. Although the concepts just described have been explored in principle by thousands of researches over the last few decades, previous efforts did not result in any HD products.