| 钱立兵* ** ***,徐进宝* ***,徐宏力**,刘济海* ***,张欢* ***.一种基于工业车辆驾驶员面部特征的疲劳检测融合方法[J].高技术通讯(中文),2025,35(4):409~418 |
| 一种基于工业车辆驾驶员面部特征的疲劳检测融合方法 |
| A fatigue detection fusion method for facial features of industrial vehicle driver |
| |
| DOI： |
| 中文关键词: 疲劳驾驶； 深度学习； 面部特征； 欧拉角； 算法部署 |
| 英文关键词: fatigue driving, deep learning, facial feature, Euler angle, algorithm deployment |
| 基金项目: |
| 作者 | 单位 | | 钱立兵* ** *** | （*安徽叉车集团有限责任公司合肥 230601）
（**中国科学技术大学计算机科学与技术学院合肥 230026）
（***安徽省工业车辆重点实验室合肥 230601） | | 徐进宝* *** | | | 徐宏力** | | | 刘济海* *** | | | 张欢* *** | |
|
| 摘要点击次数: 9 |
| 全文下载次数: 6 |
| 中文摘要: |
| 为解决工业车辆在驾驶过程中出现疲劳驾驶问题，基于驾驶员面部特征提出一种卷积神经网络的多融合疲劳检测算法。通过红绿蓝（red，green，blue，RGB）或红外摄像头捕捉驾驶员面部区域，设计轻量级人脸识别和关键点特征模型，提取面部关键特征点；根据眼部关键点提取驾驶员的眼睛框进行睁闭眼二分类模型设计，统计闭眼频率；根据嘴部关键点设计纵横比信息，统计嘴部闭合、说话及打哈欠频率；根据面部关键特征及标准三维人脸模型，计算头部姿态欧拉角信息，由头部姿态角变化范围确定注意力集中情况。算法部署在国产嵌入式低成本ARM芯片控制器上，根据融合疲劳评估算法进行语音提醒、异常时报警图像存储等。实验结果表明，白天场景下各类疲劳驾驶检测准确率达到98.3%以上，夜晚工作场景下各类疲劳驾驶检测准确率达到88.5%以上，算法检测时延控制在100ms左右，可满足工业车辆对驾驶员疲劳检测需求。 |
| 英文摘要: |
| In order to solve the problem of fatigue driving in industrial vehicles, a multi fusion fatigue detection algorithm is proposed based on driver facial features using convolutional neural networks. The driver’s face area is captured by red, green, blue (RGB) or infrared cameras, lightweight face recognition and key point feature models are designed, and key facial feature points are extracted. According to the key points of the eyes, the driver’s eye frame is extracted to design the two-classification model of the eyes, and the closing frequency is calculated. The aspect ratio information is designed based on the key points of the mouth; the frequency of mouth closure, speech and yawning is calculated. Based on the key characteristics of the face and the standard 3D face model, the Euler angle information of the head pose is computed, and the attention concentration is determined in accordance with the range of the head pose angle. The algorithm has been implemented on a domestically produced, cost-effective ARM chip controller. Based on the fusion fatigue evaluation algorithm, voice reminders are triggered and images of abnormal events are stored. Experimental results demonstrate that the accuracy rate for detecting various types of daytime driver fatigue exceeds 98.3%, while nighttime detection accuracy in working scenarios is no less than 88.5%. The algorithm’s detection delay is maintained at approximately 100ms, meeting the requirements for driver fatigue detection in industrial vehicles. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
