摘要
应用PERCLOS P80方法检测人体疲劳时,需要对人脸部,尤其是眼部的关键点进行准确而快速的定位,针对上述问题提出一种基于Retina-Face的人眼关键点检测算法。在分析Retina-Face的网络模型结构基础上,结合疲劳检测的相关场景,有针对性地对网络模型进行改进,包括:数据迁移学习、网络结构重新设计、Gabor特征提取等方法。该算法在300W数据集、自采数据集上,在检测精度和速度方面均达到较好的效果,在保证时效性的同时,提高了眼部关键点检测的准确率,也为基于计算机视觉的疲劳检测打下了基础。
When PERCLOS p80 method is used to detect human fatigue,it needs to accurately and quickly locate the key points of human face,especially the eyes.Aiming at this problem,a human eye key point detection algorithm based on retina face is proposed.Based on the analysis of retina face's network model structure,combined with the relevant scenarios of fatigue detection,the network model was improved pertinently,including data migration learning,network structure redesign,Gabor feature extraction and other methods.The algorithm has achieved good results in detection accuracy and speed on 300W data sets and self collected data sets.While ensuring timeliness,it improves the accuracy of eye key point detection,and also lays a foundation for fatigue detection based on computer vision.
作者
陈亮
郑伟
CHEN Liang;ZHENG Wei(School of Electronic Information Engineering,Beijing Jiaotong University,Beijing 100044,China;National Research Center of Railway Safety Assessment,Beijing Jiaotong University,Beijing 100044,China;Collaborative Innovation Center of Railway Traffic Safety,Beijing Jiaotong University,Beijing 100044,China)
出处
《计算机仿真》
北大核心
2023年第9期213-216,354,共5页
Computer Simulation
基金
中国国家铁路集团有限公司科技研究开发计划项目(N2021Z007)
中国铁道科学研究院集团有限公司科技研究开发计划项目(2020YJ098)
中央高校基本科研业务费专项资金资助(科技领军人才团队项目)(2022JBXT003)。
关键词
深度学习
特征金字塔
特征提取
眼部关键点
Deep learning
Feature pyramid
Feature extraction
Eye keys-points
