在“互联网+安全生产”中基于神经网络视频编码技术的研究

Research on video coding technology based on neural networks in"Internet+safety production"

5G时代的到来极大地促进了视频的应用,将视频与行业应用相结合既可以推动技术的发展,又可以通过科技改变传统模式。通过海量视频的数据进行人员行为分析、危险源识别、隐患排查等应用极大地提升了生产的安全性,实现了“互联网+安全生产”的新型管理模式。但是大量视频的传输对网络传输、存储、视频分析等带来了极大的传输和运算压力。相较于传统的视频编码方法,基于神经网络的视频编码方法可以获得更优的率失真性能,也成为当前视频编码技术领域的研究热点。文章系统梳理了基于神经网络的端到端视频编码框架,主要包括端到端P帧视频压缩和端到端B帧视频压缩。通过分析对比,P帧视频压缩方式有助于获取更高的解码视频重建质量,B帧视频压缩方式可以实现更低码率的编码传输。因此,根据5G实际的应用场景,可以灵活选择两种编码方式以满足用户不同传输带宽及不同质量的需求。

The advent of 5G era greatly promotes the application of video,and combines video andindustry application,which can not only promote the development of technology,but also change thetraditional mode through science and technology,based on the application of personnel behavioranalysis,hazard identification,hidden trouble screening and other applications greatly improves thesafety of production,and realizes the new management mode of"Internet+safe production".However,a large number of video transmission brings great transmission and computing pressure tonetwork transmission,storage,video analysis and so on.Compared with the traditional video codingmethod,the video coding method based on neural network can obtain better rate distortionperformance,which has also become one of the hot research problems in the field of video codingtechnology.This paper systematically combs the endto-end video coding framework based on neuralnetwork,which mainly includes the following two types:end-to-end P frame video compression andend-to-end B frame video compression.Through analysis and comparison,P frame video compressionis helpful to obtain higher decoding video reconstruction quality,and at the same time,Bframe videocompression can realize the encode transmission with lower code rate.Therefore,according to theactual application scenarios of 5G,we can flexibly choose two coding methods to meet the needs ofusers with diferent transmission bandwidth and diferent quality.