近年来，疲劳驾驶成为导致交通事故的重要因素之一。据统计，若驾驶员在事故发生时能快速反应半秒钟，则可阻止60%的事故发生。因此，准确检测驾驶员的疲劳状态并及时发出告警，对降低交通事故发生率以及保障驾驶员行驶安全具有重要意义。本文对基于面部特征的疲劳驾驶检测方法深入研究，主要研究工作如下：

（1）针对现有疲劳驾驶检测算法存在检测速度慢且难于部署到嵌入式设备的问题，提出了一种改进YOLOv5的疲劳驾驶检测算法。为了提升网络的特征提取能力，采用卷积模块替换主干特征提取网络中的Focus模块；算法结合疲劳驾驶图像的数据特征，对YOLOv5模型的Neck结构进行精简优化，将适合检测较大尺寸对象的19×19特征图分支剪枝，优化模型的计算效率；引入改进的CIoU函数计算回归损失，进一步提高疲劳特征的识别精度。该算法在面部特征提取的基础上，依据PERCLOS与POM参数并结合阈值对疲劳特征进行状态判定并输出结果。实验结果表明，所提算法在保障检测准确率的同时大幅降低了计算量，能较好地满足疲劳驾驶的实时检测需求。

（2）针对夜间低光照条件下图像曝光度不足导致疲劳驾驶检测准确率较低的问题，提出了一种面向夜间疲劳驾驶检测的改进Zero-DCE低光增强算法。首先采用下采样输入作为增强网络的输入，输出采用上采样映射回原始分辨率以进行图像增强，从而减少图像中的噪声干扰；其次将深度可分离卷积用于Zero-DCE模型的主干特征提取网络，降低模型参数量并提高网络的检测速度；最后引入通道一致性损失以提高图像增强效果，通过目标检测算法对面部特征进行实时检测并输出疲劳判定结果。实验结果表明，所提算法提高了面部疲劳特征的分类准确率，从而有效提高疲劳检测器在低光环境中的性能。

（3）设计并实现疲劳驾驶检测系统，将改进的YOLOv5疲劳检测算法与Zero-DCE低光增强算法部署到嵌入式设备。系统能够实时检测驾驶员疲劳状态并及时发出告警，并通过可视化大数据看板查看算法执行结果。

In recent years, fatigue driving has become one of the important factors leading to traffic accidents. According to statistics, if the driver can react quickly for half a second when an accident occurs, 60% of accidents can be prevented. Therefore, it is of great significance to accurately detect the fatigue state of the driver and issue an alarm in time to reduce the incidence of traffic accidents and ensure the driving safety of the driver. In this paper, the fatigue driving detection method based on facial features is studied in depth. The main research work is as follows:

(1) A fatigue driving detection algorithm with improved YOLOv5 is proposed to address the problems that existing fatigue driving detection algorithms have slow detection speeds and are difficult to be deploy to embedded devices. To improve the feature extraction capability of the network, the Focus module in the backbone feature extraction network is replaced by a convolutional module; the algorithm combines the data features of fatigue driving images, streamlines and optimizes the Neck structure of the YOLOv5 model, prunes the branches of the feature map suitable for detecting objects of larger size, and optimizes the computational efficiency of the model; an improved CIoU function is introduced to calculate the regression loss and The recognition accuracy of fatigue features is further improved. The algorithm is based on the extraction of facial features, the state determination of fatigue features based on PERCLOS and POM parameters and the combination of thresholds, and the output of the results. The experimental results show that the proposed algorithm can significantly reduce the computational effort while ensuring the detection accuracy, and can better meet the demand of real-time fatigue driving detection.

(2) An improved Zero-DCE low-light enhancement algorithm for nighttime fatigue driver detection is proposed to address the problem of low accuracy of fatigue driver detection due to insufficient image exposure under low-light conditions at night. Firstly, a downsampling input is used as the input to the enhancement network, and the output is upsampled and mapped back to the original resolution for image enhancement to reduce noise interference in the image; secondly, a depth-separable convolution is used for the backbone feature extraction network of the Zero-DCE model to reduce the number of model parameters and improve the detection speed of the network; Finally, a channel consistency loss is introduced to improve image enhancement, and a target detection algorithm is used to detect facial features in real-time and output fatigue determination results.The experimental results show that the proposed algorithm improves the classification accuracy of facial fatigue features, thus effectively improving the performance of the fatigue detector in low-light environments.

(3) Design and implement a fatigue driving detection system, and deploy the improved YOLOv5 fatigue detection algorithm and Zero-DCE low-light enhancement algorithm to the embedded device. The system can detect driver fatigue in real-time and issue alerts in time, and view the algorithm execution results through a visual big data dashboard.

[1]Xi J, Wang S, Ding T, et al. Detection Model on Fatigue Driving Behaviors Based on the Operating Parameters of Freight Vehicles[J]. Applied Sciences, 2021, 11(15): 32–43.

[2]Hatsukawa H, Ishikawa M. Clinical potential of pupillary light reflex parameters as objective indicators reflecting chronic rhinosinusitis-specific quality of life: a 12-month prospective longitudinal study[J]. Scientific Reports, 2021, 11(1): 21586¬21592.

[3]Zhou Z, Zhou Y, Pu Z, et al. Simulation of pedestrian behavior during the flashing green signal using a modified social force model[J]. Transportmetrica A: transport science, 2019, 15(2): 1019–1040.

[4]Zhou Z, Cai Y, Ke R, et al. A collision avoidance model for two-pedestrian groups: Considering random avoidance patterns[J]. Physica A: Statistical Mechanics and its Applications, 2017, 475: 142–154.

[5]Farah H, Hagenzieker M, Brijs T. Special issue on road safety and simulation 2017[J]. Accident analysis and prevention, 2021, 161: 106–115.

[6]Abbas Q, Alsheddy A. Driver fatigue detection systems using multi-sensors, smartphone, and cloud-based computing platforms: a comparative analysis[J]. Sensors, 2020, 21(1): 56–66.

[7]Amodio A, Ermidoro M, Maggi D, et al. Automatic detection of driver impairment based on pupillary light reflex[J]. IEEE transactions on intelligent transportation systems, 2018, 20(8): 3038-3048.

[8]Salihu A T, Hill K D, Jaberzadeh S. Neural mechanisms underlying state mental fatigue: a systematic review and activation likelihood estimation meta-analysis[J]. Reviews in the Neurosciences, 2022, 33(8): 67–80.

[9]胡习之, 黄冰瑜. 基于面部特征分析的疲劳驾驶检测方法[J]. 科学技术与工程, 2021, 21(4): 1629–1636.

[10]Chen K, Liu Z, Li Z, et al. An improved multi-source domain adaptation network for inter-subject mental fatigue detection based on DANN[J]. Biomedical Engineering, 2023(0): 12–20.

[11]王恁, 周子敬, 赵云芃, 等. 基于无线脑电信号分析的实时疲劳驾驶检测与预警研究[J]. 太原理工大学学报, 2020, 51(06): 852–859.

[12]Sikander G, Anwar S. Driver fatigue detection systems: A review[J]. IEEE Transactions on Intelligent Transportation Systems, 2018, 20(6): 2339–2352.

[13]Zhao Y, Xie K, Zou Z, et al. Intelligent recognition of fatigue and sleepiness based on InceptionV3-LSTM via multi-feature fusion[J]. Ieee Access, 2020, 8: 144205–144217.

[14]郭慧利. 基于多源信息融合的疲劳驾驶预警系统的研究[D]. 太原理工大学, 2020.

[15]李茂衡. 基于sEMG的肌肉疲劳检测方法的研究[D]. 黑龙江大学, 2020. 32(8): 147–158.

[16]李敬强, 房秋, 樊天辰, 等. 面部三维生理特征点在睡眠剥夺中的疲劳检测[J/OL]. 北京航空航天大学学报, 2023, 6: 1–12.

[17]Li X, Hong L, Wang J, et al. Fatigue driving detection model based on multi‐feature fusion and semi‐supervised active learning[J]. IET Intelligent Transport Systems, 2019, 13(9): 1401–1409.

[18]Wang Q, Mu Z. Heterogeneous signal fusion method in driving fatigue detection signals[J]. Journal of advanced transportation, 2021, 34: 1–11.

[19]蔡素贤, 杜超坎, 周思毅, 等. 基于车辆运行数据的疲劳驾驶状态检测[J]. 交通运输系统工程与信息, 2020, 20(04): 77–82.

[20]杨萍茹, 黄勇, 廖龙涛, 等. 基于智能手表的疲劳驾驶监测方法研究[J]. 重庆理工大学学报(自然科学), 2020, 34(12): 170–176.

[21]Chawla P, Rana S B, Kaur H, et al. Classification of Epileptic Seizures Using EEMD with Multi-entropy Features Integrating Different Tree Classifiers[M]//Computational Vision and Bio-Inspired Computing: Proceedings of ICCVBIC 2022. Singapore: Springer Nature Singapore, 2023: 139–157.

[22]Li K, Gong Y, Ren Z. A fatigue driving detection algorithm based on facial multi-feature fusion[J]. IEEE Access, 2020, 8: 101244–101259.

[23]马雪婷, 费树岷. 基于面部特征与深度学习的疲劳驾驶状态检测研究[J]. 电子测试, 2021, 23(11): 33–36.

[24]Reddy B, Kim Y H, Yun S, et al. Real-time driver drowsiness detection for embedded system using model compression of deep neural networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition workshops. 2017: 121–128.

[25]Tang X, Du D K, He Z, et al. Pyramidbox: A context-assisted single shot face detector[C]//Proceedings of the European conference on computer vision, 2018: 797–813.

[26]Gazda M, Hireš M, Drotár P. Multiple-fine-tuned convolutional neural networks for Parkinson’s disease diagnosis from offline handwriting[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2021, 52(1): 78–89.

[27]Zhang L, Liu F A N, Tang J. Real-time system for driver fatigue detection by RGB-D camera[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2015, 6(2): 1–17.

[28]Gao Z, Wang X, Yang Y, et al. EEG-based spatio–temporal convolutional neural network for driver fatigue evaluation[J]. IEEE transactions on neural networks and learning systems, 2019, 30(9): 2755–2763.

[29]庄员, 戚湧. 伪3D卷积神经网络与注意力机制结合的疲劳驾驶检测[J]. 中国图象图形学报, 2021, 26(01): 143–153.

[30]Sikander G, Anwar S. A novel machine vision-based 3d facial action unit identification for fatigue detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 22(5): 2730–2740.

[31]Elhaki O, Shojaei K. Output-feedback robust saturated actor–critic multi-layer neural network controller for multi-body electrically driven tractors with [formula omitted]-trailer guaranteeing prescribed output constraints[J]. Robotics and Autonomous Systems, 2022, 28(6): 154–163.

[32]赵向涛, 阳凡林, 宿殿鹏, 等. 基于弹性梯度下降模型的神经网络遥感水深反演[J]. 山东科技大学学报(自然科学版), 2022, 41(05): 1–10.

[33]Lu Z, Huang Z, Song Q, et al. An Enhanced Image Patch Tensor Decompostion for Infrared Small Target Detection[J]. Remote Sensing, 2022, 14(23): 6044–6058.

[34]Jia D, Zhou J, Zhang C. Detection of cervical cells based on improved SSD network[J]. Multimedia Tools and Applications, 2022, 81(10): 13371–13387.

[35]Jiang H, Learned-Miller E. Face detection with the faster R-CNN[C]//2017 12th IEEE international conference on automatic face & gesture recognition (FG 2017). IEEE, 2017: 650–657.

[36]许锐, 陈初侠, 乔涛. 基于MATLAB GUI的空域图像增强系统设计[J]. 电脑知识与技术, 2022, 18(04): 77–79.

[37]武鹤龙, 邱政, 张维全. 侧扫声纳图像非下采样轮廓波变换域分区增强方法[J]. 兵工学报, 2021, 42(07): 1463–1470.

[38]Lisani J L, Morel J M, Petro A B, et al. Analyzing center/surround retinex[J]. Information Sciences, 2020, 512: 741–759.

[39]Agrawal A, Jadhav N, Gaur A, et al. Improving the Accuracy of Object Detection in Low Light Conditions using Multiple Retinex Theory-based Image Enhancement Algorithms[C]//2022 Second International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT). IEEE, 2022: 1–5.

[40]Jiang Y, Gong X, Liu D, et al. Enlightengan: Deep light enhancement without paired supervision[J]. IEEE transactions on image processing, 2021, 30: 2340–2349.

[41]Wang W, Wei C, Yang W, et al. Gladnet: Low-light enhancement network with global awareness[C]//2018 13th IEEE international conference on automatic face & gesture recognition (FG 2018). IEEE, 2018: 751–755.

[42]娄平, 杨欣, 胡辑伟, 等. 基于边缘计算的疲劳驾驶检测方法[J]. 计算机工程, 2021, 47(07): 13–20.

[43]程泽, 林富生, 靳朝, 等. 基于轻量化卷积神经网络的疲劳驾驶检测[J]. 重庆理工大学学报(自然科学), 2022, 36(02): 142–150.

[44]李昭慧, 张玮良. 基于改进YOLOv4算法的疲劳驾驶检测[J]. 电子测量技术, 2021, 44(13): 73–78.

[45]于硕, 李慧, 桂方俊, 等. 复杂场景下基于YOLOv5的口罩佩戴实时检测算法研究[J]. 计算机测量与控制, 2021, 29(12): 188–194.

[46]Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[C]//Proceedings of the AAAI conference on artificial intelligence. 2020, 34(07): 12993–13000.

[47]Yue X, Li H, Shimizu M, et al. Deep learning-based real-time object detection for empty-dish recycling robot[C]//2022 13th Asian Control Conference (ASCC). IEEE, 2022: 2177–2182.

[48]赵睿, 刘辉, 刘沛霖, 等. 基于改进YOLOv5s的安全帽检测算法[J/OL]. 北京航空航天大学学报, 2023: 1–16.

[49]Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition, 2017: 2117–2125.

[50]Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[C]//Proceedings of the AAAI conference on artificial intelligence, 2020, 34(07): 12993–13000.

[51]Amori K E, Abd-AlRaheem M A. Field study of various air based photovoltaic/thermal hybrid solar collectors[J]. Renewable Energy, 2014, 63: 402–414.

[52]汪洪涛, 谢牡丹, 潘昊. 基于面部特征的疲劳驾驶检测方法研究[J]. 武汉理工大学学报(交通科学与工程版), 2021, 45(05): 851–862.

[53]Abtahi S, Omidyeganeh M, Shirmohammadi S, et al. YawDD: A yawning detection dataset[C]//Proceedings of the 5th ACM multimedia systems conference, 2014: 24–28.

[54]黄佳琦, 范军芳, 李蓓蓓. CenterNet-based遥感图像目标检测方法研究[J]. 弹箭与制导学报, 2023, 43(01): 24–40.

[55]Yuan W. Accuracy comparison of YOLOv7 and YOLOv4 regarding image annotation quality for apple flower bud classification[J]. AgriEngineering, 2023, 5(1): 413–424.

[56]Li C, Guo C, Loy C C. Learning to enhance low-light image via zero-reference deep curve estimation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(8): 4225–4238.

[57]Zhang J, Craigmile P F, Cressie N. Loss function approaches to predict a spatial quantile and its exceedance region[J]. Technometrics, 2018, 50(2): 216–227.

[58]Li Z. Remote Sensing; Studies Conducted at National Central University on Remote Sensing Recently Published[J]. Journal of Technology, 2020, 124(1): 75–88.

[59]Nieves J L, Pérez-Ocón F, Hernández-Andrés J, et al. Spectral-reflectance function recovery for improved colour-constancy experiments[J]. Displays, 2022, 23(5): 213–222.

[60]Berloffa G, Modena F. Income shocks, coping strategies, and consumption smoothing: An application to Indonesian data[J]. Journal of Asian Economics, 2013, 24: 158–171.

[61]杨学存, 和沛栋, 陈丽媛, 等. 基于深度可分离卷积的轻量级YOLOv3输电线路鸟巢检测方法[J]. 智慧电力, 2021, 49(12): 88–95.

[62]Kim J, Lee J K, Lee K M. Deeply-recursive convolutional network for image super-resolution[C]//Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 1637–1645.

[63]Devi G. Medicinal plant: vitex negundo[J]. International Journal of Current Research, 2021, 13: 17592–17594.

[64]Li M, Liu J, Yang W, et al. Structure-revealing low-light image enhancement via robust retinex model[J]. IEEE Transactions on Image Processing, 2018, 27(6): 2828–2841.

[65]Richard J. Flip flops: Digital material of failed musicals can offer salutary lessons[J]. The Stage, 2022, 11(3): 88–93.