2018 年 12 月 Chinese Journal on Internet of Things December 2018
第 2 卷第 4 期 物 联 网 学 报 Vo l .2
No.4
物联网环境下基于场景要素的多码流变分辨率压缩传输技术研究
肖尚武,胡瑞敏,陈宇,肖晶
(武汉大学计算机学院国家多媒体软件工程技术研究中心,湖北 武汉 430072)
摘 要:针对城市监控覆盖面广、海量接入的需求,实现低带宽和低功耗性能是解决这一问题的重要研究方向。
在智慧城市、安防监控等应用领域,基于场景要素,如人脸关键区域的视频监控尤为重要。实现场景要素的提取,
以极低带宽传输关键信息,通过多码流区别编码策略,在物联网环境下实现视频技术的应用,是目前值得研究的
可行方向。通过设计面向人脸的变分辨率混合编码算法,可大幅度节省带宽、降低功耗,满足窄带物联网的接入
要求。通过基于深度学习 Caffe 框架的人脸检测算法,在关键帧获取人脸感兴趣区域,并以高分辨率编码人脸图
像;通过设计码率自适应分配算法,合理利用带宽,区别编码人脸信息和全图背景内容;通过窄带传输编码后的
混合码流信息,在接收端采用基于关键帧的人脸增强解码算法,得到人脸局部的高清监控画面。实验表明,采用所
提方法在 120~160 kbit/s 窄带传输时,人脸画面可以保持与原始高清监控采集端同等清晰度,具有很强的实用性。
关键词:NB-IoT;监控视频;变分辨率;视频编码;人脸检测
中图分类号:TP37
文献标识码:A
doi: 10.11959/j.issn.2096-3750.2018.00075
Research on multi-stream variable resolution compression and
transmission technology based on scene elements in
Internet of things environment
XIAO Shangwu, HU Ruimin, CHEN Yu, XIAO Jing
National Engineering Research Center for Multimedia Software, Wuhan University, Wuhan 430072, China
Abstract: In response to the demand of wide coverage and massive access, low bandwidth and low power consumption
is an important research direction to solve this problem. In smart cities, security monitoring and other application areas,
video surveillance based on the region of interest of the face are particularly important. It is a feasible direction to realize
the extraction of scene elements, transmission of key information with very low bandwidth and the application of video
technology in the Internet of things environment through the strategy of multi-stream differential coding. By designing a
face-oriented variable resolution hybrid coding algorithm, the bandwidth could be saved and the power consumption
could be reduced greatly, the access requirements of narrowband Internet of things could be met. Through the face detec-
tion algorithm based on the deep learning Caffe framework, the face region of interest was acquired in key frames, and
the face image was encoded with high resolution. By designing the code rate adaptive allocation algorithm, the bandwidth
was utilized rationally, and the encoded face information and the full background content were distinguished. The encod-
ed mixed code stream information was transmitted through the narrowband; the key frame-based face enhancement de-
coding algorithm was adopted at the receiving end to obtain a partial HD high-definition monitoring picture. Experiments
show that when the video encoded by the proposed method is transmitted in a narrow band whose transmission rate is
120~160 kbit/s, the face image can maintain the same definition as the original HD monitoring acquisition end, which has
strong practicability.
Key words: NB-IoT, surveillance video, variable resolution, video coding, face detection
收稿日期
:2018-08-09;修回日期:2018-10-29