基于自然驾驶工况的驾驶员变道意图探测方法研究  

Study of Drivers L ane Change Intention Detection Method based on Naturalistic Driving

作　　者：周鹏飞[1] 朱西产[1] 刘兴亮[1] 马志雄[1] 王大志[2] 刘军勇[2] 

机构地区：[1]同济大学汽车学院,上海201804 [2]上海汽车集团股份有限公司技术中心,上海201804

出　　处：《上海汽车》2016年第6期36-42,共7页Shanghai Auto

基　　金：上海市科学技术委员会资助项目(编号:13QB1402300)

摘　　要：通过对自然驾驶工况进行车道变换的筛选,共收集到349个车道保持和变道意图样本。利用Filter方法对变道意图参数进行特征选择,得到方向盘转角速率、车辆横摆角和方向盘转角等3个特征参数。然后建立基于高斯隐马尔可夫理论的变道意图识别模型,其中在高斯混合度M为8时,达到了87.25%的变道识别准确率,并且模型的受试者工作特性较理想。最后对模型进行了验证分析,表明该模型能够为高级驾驶辅助提供支持,从而提高驾驶安全性。By selecting lane change scenarios from naturalistic driving conditions, 349 lane keeping and lane change samples are collected. Then using Filter method to select feature parameters related to lane change, it obtained three main feature parameters： steering wheel angle rate, vehicle yaw rate and steering wheel angle. After establishing lane change intention detection model based on Gaussian Mixture Hidden Markov Model theory, when the Gaussian mix parameter M equals 8, the GM - HMM model reaches 87.25% recognition accuracy and the receiver operating characteristics of the model works well too. Finally, the model is validated and the results show that the intention recognition model can provide decision - making support for the advanced driving assistance system, which is helpful to improve the safety of driving.

关 键 词：车道变换 意图探测 特征选择 高斯隐马尔可夫模型 

分 类 号：U463.6[机械工程—车辆工程]