基于EKF学习方法的BP神经网络汽车换道意图识别模型研究

摘要

Timely and accurate recognition of lane change intention is useful to improve the vehicles' safety,leading to Safety Driving Assistant.A model based on BP (Back-propagation) Neural Network with extended Kalman learning method was developed to recognize the lane change intention,and predict the next behavior.Twenty naturalistic driving experiments were taken to acquire the time-headway,steering angle,driver's head motion and lane departure.Data from the twenty on-road experiments verifies the validation of this model.The results shows that the proposed model has a higher accuracy,shorter recognition time and better robust performance than the Hidden Markov Method,Dynamic Bayesian Method before.Driving in the lane change and car follow situations,the model can successfully recognize the lane change intention at an accuracy of 95 percentages.%实时准确地识别驾驶人的换道意图有助于提高车辆行驶的安全性,达到安全辅助驾驶的目的.文中提出了一种基于EKF(extended kalman filter)学习方法的BP神经网络模型,用于识别驾驶人的换道意图,并进行短时行为预测.通过实验采集了20组高速公路实车行驶数据,利用前方车头时距、转向盘转角值、驾驶人头部水平位置数据,以及车道偏离量4类数据样本进行训练得到结果.实验结果表明:本模型较传统的神经网络识别模型具有更短的识别时间,且模型的可信度更高.在车辆换道和直线行驶2种工况下,本模型对换道意图的识别准确率达到了95％.