基于车辆横向运行数据的遗传-支持向量机算法的分心驾驶状态判别模型  

GA-SVM Distracted Driving State Discrimination Model Based on Vehicle Lateral Running Data

作　　者：文香 邓超 WEN Xiang;DENG Chao(School of Automobile and Traffic Engineering,Wuhan University of Science and Technology,Wuhan 430065,China;Intelligent Automobile Engineering Research Institute,Wuhan University of Science and Technology,Wuhan 430065,China;Unmanned System Intelligent Perception Control Technology Engineering Laboratory of Sichuan Province,Chengdu 610225,China;Laboratory of Cloud IoT Intelligent Equipment in Expressway Construction and Maintenance,Jinan 250357,China)

机构地区：[1]武汉科技大学汽车与交通工程学院,武汉430065 [2]武汉科技大学智能汽车工程研究院,武汉430065 [3]四川省无人系统智能感知控制技术工程实验室,成都610225 [4]云基物联网高速公路建养设备智能化实验室,济南250357

出　　处：《科学技术与工程》2023年第25期10990-10996,共7页Science Technology and Engineering

基　　金：国家自然科学基金青年科学基金(52002298);教育部产学合作协同育人项目(202102580026);四川省无人系统智能感知控制技术工程实验室开放课题(WRXT2022-001);云基物联网高速公路建养设备智能化实验室开放课题(KF_2022_301002);“运输车辆检测、诊断与维修技术”交通行业重点实验室开放课题(JTZL2205);武昌工学院科学研究项目(2022KY24)。

摘　　要：分心状态是造成交通事故的重要原因。当前侵入式与半侵入式检测多被应用于分心驾驶识别,此方法会对驾驶任务产生一定干扰,且成本相对较高。对此提出一种低成本的基于车辆横向运行数据的分心驾驶状态判别方法。实验选取手机通话作为分心影响因素,设计了正常驾驶、免提通话、手提通话3个维度。首先,基于驾驶模拟器采集的数据,对正常与分心状态下的车辆运行指标进行Man Whitney U检验,从时域及频域中提取出与分心驾驶显著相关的车辆横向控制指标;其次,构建支持向量机(support vector machine,SVM)分心状态判别模型,将径向基函数作为SVM的核函数,使用网格搜索算法(grid search algorithm,GSA)、粒子群算法(particle swarm optimization,PSO)及遗传算法(genetic algorithm,GA)对SVM模型参数进行优化;最后,对比GSA-SVM、PSO-SVM与GA-SVM分心判别模型的分类效果,并运用接受者操作特征(receiver operating characteristics,ROC)曲线对模型性能进行评估。研究结果表明:GA-SVM分心判别模型的最优交叉验证率、准确率及F1分别为87.9%、91.9%、94.05%,高于GSA-SVM(86.2%、87.2%、90.35%)与PSO-SVM(87.9%、91.2%、93.46%);GA-SVM判别模型ROC曲线接近于(0,1)坐标,其曲线下的面积(area under curve,AUC)为0.9353。所提出的GA-SVM分心判别模型获得了较好的分类效果,故认为此模型适合作为低成本条件下基于车辆横向控制的分心驾驶状态判别模型。Distraction is an important cause of traffic accidents.Currently,invasive and semi-invasive detection are mostly used for distracted driving recognition,but this method cause certain disturbances to driving task,and the cost is relatively high.Therefore,a low-cost method of discriminating the distracted driving state based on vehicle lateral running data was proposed.Mobile phone calls were selected as the distraction influencing factor,and three dimensions of normal driving,hands-free calling,and hand-held calling were designed.Firstly,based on the vehicle operation data collected by the driving simulator,the Man Whitney U test was performed on the indicators in normal and distracted states,and the vehicle lateral control indicators significantly related to distracted driving were extracted from the time domain and frequency domain.Secondly,support vector machine(SVM)distraction state discrimination model was built,and the grid search algorithm(GSA),particle swarm optimization algorithm(PSO)and genetic algorithm(GA)were used to optimize the parameters of the SVM model.Finally,the classification effects of GSA-SVM,PSO-SVM and GA-SVM distraction discriminant models were compared,and the receiver operating characteristics(ROC)curve was used to evaluate the performance of the models.The research results show that the optimal cross-validation rate,accuracy and F1 of the GA-SVM distraction discriminant model are 87.9%,91.9%,and 94.05%,respectively,which are higher than those of GSA-SVM(86.2%,87.2%,90.35%)and PSO-SVM(87.9%,91.2%,93.46%).The ROC curve of the GA-SVM discriminant model is close to the coordinate of(0,1),and its area under curve(AUC)is 0.9353.The proposed GA-SVM distraction discrimination model achieves good classification effect and verifies its validity.Therefore,it is considered that this model is suitable for the distracted driving state discrimination model based on vehicle lateral control under low-cost conditions.

关 键 词：交通安全 分心驾驶 特征提取 支持向量机 参数寻优 

分 类 号：U461[机械工程—车辆工程]