基于对比学习的驾驶员异常驾驶行为检测算法

打开文本图片集

中图分类号：TP391.41 文献标志码：A DOI： 10.20104/j.cnki.1674-6546.20240304

【Abstract】In the processof driving a vehicle，the complex and changing environment inside the vehicle，the change of lightingconditionsandthediversityofdrivers’behavioralposturesaffectthedetectionandrecognitionofabnormaldriver behavior.Toaddressthisisue，thispaper proposesadriverabnormaldriving behaviordetectionalgorithmbasedoncontrast learning.Thepaperfirstlyconsiders driver’sdriving behaviordetectionasabinaryclassficationtask，andutilizesacontrast learningapproach tocompare driver’snormaldriving withabnormal driving samplesandto improve the performanceof the modelbycontrastinglossfunctions.Secondlythedepthimagesrightaheadandabovethedriverservesasinputstosolvethe problemsofcomplex in-vehicleenvironmenttochangethelight intensityand blindspots inviewpointbyproviding thedepth informationofthedriver.Finaly3DconvolutionisintroducedinthelightweightnetworkMobileNetV2，andeoratioof channel blending isaddedtotheconvolutionlayerofeachbotleneck structuretoimprovetheauracyofrecognition.Test results show that accuracy of the proposed algorithm reaches 94.18% in the Driver’s Abnormality Detection （DAD） dataset and ROC AUC reaches O.962，which shows theefectivenessof the algorithm indriver’sabnormal behaviordetection.

Keywords：Abnormal driving behavior detection，Contrast learning，Second classification，3D Convolution Neural Networks（CNN）

【引用格式】李仲伦，于光达，杨帅，等.基于对比学习的驾驶员异常驾驶行为检测算法[J].汽车工程师，2025（8）：29-36. LI ZL，YUGD，YANG S，etal.Driver Abnormal Driving Behavior Detection Algorithm Basedon Contrast Learning[J]. Automotive Engineer， 2025（8）： 29-36.

1前言

HighwayTraffic SafetyAdministration，NHTSA）的数据， 80% 的交通事故和 16% 的公路死亡是由驾驶员分心驾驶引起的。（剩余12456字）