摘要
针对疲劳驾驶检测要求精度高、实时性、鲁棒性等特点,提出一种基于轻量化网络的疲劳驾驶检测模型.采用Faceboxes进行人脸定位,通过PFLD(praclical facial landmark detector)检测人脸关键点以获取眼部、嘴部区域图像和头部姿态;然后,基于Xception的模块化思想,设计眼、嘴状态分类网络,准确率分别达到99.61%、97.58%;最后,分别计算基于时间序列的眼部、嘴部及头部疲劳表征参数,采用支持向量机(Support Vector Machine,SVM)进行疲劳判定.经实验表明:疲劳检测准确率达到98.3%,速率为47 fp/s,模型大小为7.1 Mb.在保留整个模型轻量化的同时,兼顾准确率和实时性,可应用在嵌入式系统或低算力设备中.
Aiming at the characteristics of high precision,real-time ness and robustness of fatigue driving detection,this paper proposes a fatigue driving detection model based on lightweight network.Faceboxes face location is adopted,and face key points are detected by PFLD(Practical Facial Landmark Detector)model to obtain eye and mouth images and head pose;then,based on the modular idea of Xception,the eye and mouth state classification network is designed,and the accuracy rates are 99.61%and 97.58%respectively;finally,the paper calculate the fatigue characterization parameters of the eyes,mouth and head based on time series respectively,and establishes a support vector machine model for fatigue determination.The experiment result shows that the accuracy of fatigue detection is 98.3%,the speed is 47 fp/s,and the size of the model is 7.1 M.It can be used in embedded systems or low computing power devices while preserving the lightweight of the whole model and taking into account the accuracy and real-time ness.
作者
刘志英
武斌
LIU Zhiying;WU Bin(School of Computer and Information Engineering,Tianjin Chengjian University,Tianjin 300384,China)
出处
《天津城建大学学报》
CAS
2023年第1期49-54,共6页
Journal of Tianjin Chengjian University
基金
国家自然科学基金资助项目(61902273).
