【摘要】 为更准确地对驾驶行为进行预警,进一步提高驾驶人换道意图的辨识准确率,借助驾驶模拟器采集数据,建立基于支持向量机(SVM)理论的换道意图辨识模型。对比分析不同人-车-路系统参数组合在换道意图和车道保持期间的差异性,选取最佳特征参数组合,运用网格和遗传算法-支持向量机(GA-SVM)寻优方法优化模型参数,并对优化模型进行验证。结果表明,以纵向加速度、方向盘转角、车辆偏离车道中心线的距离、驾驶人头部运动横坐标变化值作为表征换道意图的人-车-路系统特征参数,优化模型惩罚参数c为58.642 3、核函数参数g为222.732 6时,该模型对驾驶人换道意图的辨识准确率为90%,误警率为5%,基本实现准确辨识换道意图。更多还原

【Abstract】 With the aid of the driving data collected from a driving simulator,a SVM lane change intention identification model was built in order that accurate early warning on driving behavior can be made and then accuracy of the lane change intention identification can be improved. By comparing differences in characteristic parameters of a human-vehicle-road system while at lane changing and lane keeping,the optimized characteristic parameters were determined. The model parameters were optimized by using a grid optimization method combined with a genetic algorithm-SVM( GA-SVM) method,and then the optimization model was verified. In a human-vehicle-road system,the longitudinal acceleration,the rotating angle of steering wheel,the departure distance off the centerline and the horizontal value of the driver ’s head were taken as characteristic parameters for the lane change intention identification. When the penalty parameter c is 58. 642 3 and the kernel function parameter g is 222. 732 6,the accuracy of lane change intention identification is 90%,while the rate of false alarm is 5%. In summary,the lane change intention can be accurately identified.更多还原