面向实时视频流分析的边缘计算技术

Edge computing technologies for streaming video analytics

实时视频流分析在智能监控、智慧城市、自动驾驶等场景中具有重要价值.然而计算负载高、带宽需求大、延迟要求严等特点使得实时视频流分析难以通过传统的云计算范式进行部署.近年来兴起的边缘计算范式,将计算任务从云端下沉到位于网络边缘的终端设备和边缘服务器上,能够有效解决上述问题.因此,许多针对实时视频流分析的边缘计算研究逐渐涌现.本文首先介绍了智能视频流分析和边缘计算的背景知识,以及二者结合的典型应用场景;接着提出了现有系统所关注的衡量指标和面临的挑战;然后从终端设备层次、协作层次、边缘/云层次对本领域的关键技术分别进行了详细的介绍,重点涉及了模型压缩和选择、本地缓存、视频帧过滤、任务卸载、网络协议、隐私保护、查询优化、推理加速和边缘缓存技术.基于对上述各项核心技术的有机整合,本文提出了基于边缘计算的视频大数据智能分析平台Argus,从数据采集、推理分析,到数据挖掘、日志管理,对实时视频流分析全生命周期提供支持,并成功应用到智慧油田中.最后,本文讨论了本领域尚待解决的问题和未来研究方向,希望为今后的研究工作提供有益参考.

Real-time streaming video analytics is important in applications such as intelligent surveillance, smart city, and autonomous driving. However, large-scale streaming video analytics is impractical on the cloud, due to its high computation demand, large bandwidth consumption and stringent latency requirement. The emerging edge computing paradigm can effectively solve these problems by pushing computation from the cloud to devices and servers at the network edge. To this end, this article conducts a comprehensive survey on edge computing technologies for real-time streaming video analytics. Firstly, it introduces the background of video analytics and edge computing, as well as the typical application of edge-based steaming video analytics. Then, it proposes the performance indicators and challenges faced by the existing systems. Afterwards, the key technologies in this field are introduced in detail from the device level, collaboration level, and edge/cloud server level, including model compression and selection, local caching, frame filtering, task offloading, streaming protocol, privacy protection,query optimization, inference acceleration, and edge caching. Based on the integration of above core technologies,this article proposes an edge-based large-scale video analytics platform, Argus, which provides systematic support for the real-time video stream analytics on video collection, model inference, data mining, and log management.Argus has been successfully deployed in the smart oilfield scenario. Last but not least, this article discusses the open issues on edge-based streaming video analytics, in the hope of inspiring future research ideas.