摘要
文章实现了基于Yolov5s和Dlib的视频实时人脸识别。基于开源的人脸数据集CelebA重新训练Yolov5s网络,使得Yolov5s能够检测并标记出图像中的人脸位置。将Yolov5s检测到的人脸区域图像输入到第三方人脸识别模块Dlib中,首先提取出68点人脸面部关键点,再将关键点生成人脸特征向量,同时通过Dlib模块提取事先准备好的需要识别的人脸图像的特征向量,并保存到相应的人脸数据库中。由于同一个人的人脸特征向量映射到高维空间的距离是接近的,基于此引入更高效Annoy算法对人脸特征向量创建索引,提高人脸识别速度。
This paper realizes video real-time face recognition based on yolov5s and Dlib.yolov5s network is retrained based on the open source face data set celebA,so that yolov5s can detect and mark the face position in the image.The face region image detected by yolov5s is input into the third-party face recognition module Dlib.Firstly,68 face key points are extracted,and then the key points are generated into face feature vector.At the same time,the feature vector of the prepared face image to be recognized is extracted through the Dlib module and stored in the corresponding face data database.Because the distance between the face feature vector of the same person and the high-dimensional space is close,based on this,a more efficient annoy algorithm is introduced to index the face feature vector to improve the speed of face recognition.
作者
黄振龙
吴林煌
HUANG Zhen-long;WU Lin-huang(College of Physics and Information Engineering,Fuzhou University,Fuzhou 350108,China)
出处
《电脑知识与技术》
2021年第32期94-96,共3页
Computer Knowledge and Technology
作者简介
黄振龙(1999-),男,福建福州人,硕士,主要研究方向为多媒体技术;吴林煌(1984-),男,福建漳州人,博士,副研究员,主要研究方向为AI智能信号处理、视频编码、FPGA与嵌入式系统设计。
