基于瑞芯微平台的轻量化人脸属性估计视频流设计方法  

A Lightweight Video Streaming Design Method for Face Attribute Estimation Based on RockChip Platform

作　　者：丁鑫 谢婷婷 巴梓轩 DING Xin;XIE Ting-ting;BA Zi-Xuan(The 27th Research Institute of China Electronics Technology Group Corporation,Zhengzhou 450047,China)

机构地区：[1]中国电子科技集团公司第二十七研究所,郑州450047

出　　处：《电光系统》2024年第2期19-23,28,共6页Electronic and Electro-optical Systems

摘　　要：年龄估计系统能够自动判断人的年龄。年龄信息作为一种生物特征,除了对使用者身份有着非常重要的影响外,还大大提高人脸识别系统的性能。随着深度学习方法的兴起,人脸识别的深度学习算法的识别率和准确率也越来越高,但是人脸和性别年龄等特征的综合识别和评估的实时性不足,基于深度学习的轻量化设计及优化成为主流。开展基于瑞芯微RK3588平台的轻量化人脸年龄估计的研究,采用主流的YOLOv8端到段模型进行人脸检测,对检测出来的人脸进行性别和年龄的估计,并利用H264将视频进行编码,通过RTSP(RealTimeStreamingProtocol)实时流协议进行视频流推送到远端进行显示,使用widerface数据集结果测试显示,优化后的模型精度降低2.5%左右,速度提高1倍,编码速度提高了10倍,实现了人脸年龄估计视频流的推送。Age estimation system can automatically determine the age of a person.Age information as a biometric feature has a very important impact on the user's identity,and also greatly improves the performance of the face recognition system.With the rise of deep learning methods,the recognition rate and accuracy of deep learning algorithms for face recognition is also increasingly high,but the real-time of the comprehensive recognition and evaluation of the features of the face and the gender and age is insufficient.Lightweight design and optimization based on the deep learning has become a mainstream.This paper carries out research on lightweight face age es-timation based on RK3588 platform,uses mainstream YOLOv8 end-to-segment model for face detection,esti-mates the gender and age of the detected face,and uses H264 to encode the video to be pushed to the remote end for display via RTSP video stream,using widerface dataset results test.It shows that the accuracy of the op-timized model is reduced by about 2.5%and the speed is increased by 1x,with 10 times faster encoding,which realizes the push of the face age estimation video stream.

关 键 词：人脸识别 年龄估计 RK3588 视频流 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]