基于BiLSTM的多模态矿车疲劳驾驶检测方法

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）24-0027-06

Abstract：To addressthe fatigue driving issue caused bylong-hour operations of mining vehicle drivers，a study is conducted onamultimodal fatigue detection model basedon YuNet-SLPT-ResNet-BiLSTM（YSRB）.The model constructs anend-to-endsystem，wherefacedetectionisachievedthroughtheYuNetnetwork，98-facekeypointsarelocalizedcombining theSLPTmodel，head poseanglesaredetectedviaResNet，andtemporal featuresare modeled trough theBiLSTMnetwork. Byfusing mutimodalfeatures includingfacialcharacteristics，headposes，andtemporaldynamics，themodelconductsfatigue statedetermination.Experimentalresultsontheself-builtDMS-7datasetshowthatthe modelachievesadetectionaccuracyof 98.41% ，with precision and recall rates reaching 98.2% and 98.1% respectively， significantly outperforming traditional methods. This approach exhibits good robustness andcan efectively handle complex conditions such asvehicle vibrations and varying lighting n mine environments，providing areliable solution for fatigue driving detection in complex mine environments.

Keywords： fatigue driving detection; face detection; BiLSTM; multi-task learning; mine safety

0 引言

随着矿山开采规模的不断扩大，露天矿卡数量逐年增加，截至2024年，我国露天煤矿矿车数量约30000辆[1]。（剩余11057字）