Several limitations of commonly used modalities, including visual cues, physiological signals, and vehicle behavior, are encountered in real-world driver fatigue detection scenarios. Valuable information related to the driver's fatigue state is also contained in speech signals. To improve the accuracy and robustness of fatigue detection methods, a lightweight multimodal approach integrating audio and visual modalities was proposed. This method was designed to leverage the complementary strengths of acoustic and ocular features for more precise and reliable driver fatigue detection. For the audio modality, the CNN-SE-MN4 model was proposed by incorporating the efficient and lightweight MobileNet v4 architecture into a mainstream CNN-SE network, enabling fatigue assessment based on spectrogram analysis. For the visual modality, the YOLOv8n-MN4 model was developed by replacing the backbone of YOLOv8n with MobileNet v4 and introducing a parameter-free attention mechanism into the neck network, facilitating the detection of the driver's eye region and subsequent fatigue state inference from eye images. Finally, weights were assigned to fatigue indicators from both modalities through fuzzy analytic hierarchy process, enabling comprehensive fatigue level evaluation. The results show that the improvements made to mainstream models using MobileNet v4 are all effective. Compared with the traditional equal-weight fusion method, the F1-score is improved by 5.7 percentage points, confirming the effectiveness of incorporating the audio modality.
LIN Hao, LI Lei-xiao.
Lightweight Multimodal Fatigue Driving Detection Based on Integrating Audio and Vision Modalities[J].
Science Technology and Engineering, 2026, 26(22): 9586-9595 DOI:10.12404/j.issn.1671-1815.2506963
(4)端点检测。由于人在驾驶的环境中,不会持续发出声音,需要从连续的音频中分离出有效的声音部分。端点检测就是利用语音信号的短时能量(short-time energy,STE)和短时过零率(short-time zero crossing rate,STZCR)这两个特征来检测语音的起始和结束点。这种方法简单、计算量小,适用于实时语音处理系统。
利用眼部图像进行疲劳判定,通常采取闭眼时长和眨眼的频率进行计算,即PERCLOS(percentage of eyelid closure over the pupil over time)。PERCLOS是利用一定时间内驾驶人的眼睛闭合程度超过一定阈值的时间百分比来判定驾驶人是否疲劳。采用P80[24]判断疲劳,即眼睑闭合到覆盖瞳孔的80%时的状态视为疲劳,如图7所示。
YangYanyan, LiLeixiao, LinHao. Review of research on fatigue driving detection based on driver facial features[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(6): 1249-1267.
YangJucheng, WeiFeng, LinLiang, et al. A research survey of driver drowsiness driving detection[J]. Journal of Shandong University(Engineering Science Edition), 2024, 54(2): 1-12.
[7]
ZhaoG, HeY, YangH, et al. Research on fatigue detection based on visual features[J]. IET Image Processing, 2022, 16(4): 1044-1053.
[8]
YiY, ZhouZ, ZhangW, et al. Fatigue detection algorithm based on eye multi-feature fusion[J]. IEEE Sensors Journal, 2023, 23(7): 7949-7955.
[9]
SunZ, MiaoY, JeonJ Y, et al. Facial feature fusion convolutional neural network for driver fatigue detection[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106981.
[10]
JiaH, XiaoZ, JiP. Real-time fatigue driving detection system based on multi-module fusion[J]. Computers & Graphics, 2022, 108: 22-33.
[11]
LiX, LinH, DuJ, et al. Computer vision-based driver fatigue detection framework with personalization thresholdand multi-feature fusion[J]. Signal, Image and Video Processing, 2024, 18(1): 505-514.
[12]
DuG, LiT, LiC, et al. Vision-based fatigue driving recognition method integrating heart rate and facial features[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 22(5): 3089-3100.
[13]
DuG, ZhangL, SuK, et al. A multimodal fusion fatigue driving detection method based on heart rate and PERCLOS[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 21810-21820.
[14]
LiuK, FengG, JiangX, et al. A feature fusion method for driving fatigue of shield machine drivers based on multiple physiological signals and auto-encoder[J]. Sustainability, 2023, 15(12): 9405.
[15]
HeC, XuP, PeiX, et al. Fatigue atthe wheel: a non-visual approach to truck driver fatigue detection by multi-feature fusion[J]. Accident Analysis & Prevention, 2024, 199: 107511.
ZhuJinglong, LiuQiang, LiBo, et al. Driver fatigue state discrimination based on respiratory fatigue node[J]. Science Technology and Engineering, 2024, 24(16): 6927-6934.
ZhangBingtao, YanDachuan, WangXiaomin, et al. Fatigue driving detection based on fusion of EEG and blink potential features[J]. Journal of Railway Science and Engineering, 2025, 22(8): 3738-3747.
LiTaiguo, ZhouXinghong, LiQuanqin, et al. Train driver fatigue detection based on UMAP and speech multi-feature fusion[J]. Journal of Railway Science and Engineering, 2024, 21(10): 4014-4026.
YangChangqi, ZhangYuxuan. Controller fatigue recognition based on speech features[J]. Science Technology and Engineering, 2025, 25(29): 12732-12738.
[26]
RiadR, DenaisM, de GennesM, et al. Automated speech analysis for risk detection of depression, anxiety, insomnia, and fatigue: algorithm development and validation study[J]. Journal of Medical Internet Research, 2024, 26: e58572.
[27]
AlikaşifoğluT, KartalB, KoçA. Wiener filteringin joint time-vertex fractional fourier domains[J]. IEEE Signal Processing Letters, 2024, 31: 1319-1323.
[28]
DongZ, WangD, YinJ, et al. The CNN-SE-LSTM model for phase interferometer ambiguity resolution[J]. IEEE Signal Processing Letters, 2025, 32: 2639-2643.
[29]
MaW, SongG, ZengQ, et al. FFCSLT: a deep learning modelfor traffic police hand gesture recognition using surface electromyographic signals[J]. IEEE Sensors Journal, 2024, 24(8): 13640-13655.
[30]
QinD, LeichnerC, DelakisM, et al. MobileNetv4: universal models for the mobile ecosystem[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2024: 78-96.
[31]
YangL, ZhangR Y, LiL, et al. Simam:a simple, parameter-free attention module for convolutional neural networks[C]// International Conference on Machine Learning. Online: PMLR, 2021: 11863-11874.
PanZhigeng, LiuRongfei, ZhangMingmin. Research on fatigue driving detection algorithm based on fuzzy comprehensive evaluation[J]. Journal of Software, 2019, 30(10): 2954-2963.
[34]
MoslemS. A novel parsimonious spherical fuzzy analytic hierarchy processfor sustainable urban transport solutions[J]. Engineering Applications of Artificial Intelligence, 2024, 128: 107447.
[35]
CebiS, GündoğduF K, KahramanC. Consideration of reciprocal judgments through decomposed fuzzy analytical hierarchy process: a case studyin the pharmaceutical industry[J]. Applied Soft Computing, 2023, 134: 110000.
WangShuguo, LiLeixiao, ZhaoXinying. A fatigue driving detection method based on multiple lateral states of drivers[J]. Journal of Inner Mongolia University of Technology(Natural Science Edition), 2025, 44(6): 542-551.
WangHongjun, BaiHao, ZhaoHui, et al. Driver fatigue state detection and early warning technology based on computer vision[J]. Science Technology and Engineering, 2022, 22(12): 4887-4894.
[40]
HuJ, ShenL, SunG. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Salt Lake City: IEEE, 2018: 7132-7141.
[41]
LiuY, ShaoZ, HoffmannN. Global attention mechanism: retain information to enhance channel-spatial interactions[J]. arXiv Preprint arXiv: 2112.05561, 2021.
[42]
RuanD, WangD, ZhengY, et al. Gaussian context transformer[C]// Proceedings of the IEEE/CVF CVPR. Online: IEEE, 2021: 15129-15138.
[43]
WengC H, LaiY H, LaiS H. Driver drowsiness detection via a hierarchical temporal deep belief network[C]// Asian Conference on Computer Vision. Cham: Springer International Publishing, 2016: 117-133.
[44]
ChenP, JianQ, WuP, et al. A multi-domain fusion human motion recognition method based on lightweight network[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 1-5.
[45]
HuangR, WangY, LiZ, et al. RF-DCM: multi-granularity deep convolutional model based on feature recalibration and fusion for driver fatigue detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 23(1): 630-640.