辅助驾驶中的换道决策安全研究

换道是驾驶员达到快速通行目标的一种常用手段,但换道会带来很多公路交通事故。为有效避免交通事故,需给驾驶员提供换道安全预警。构建了安全换道决策模型,将换道决策分为换道意图识别和换道条件判断分别建立模型以提高预测精确度。通过神经网络方法SOM(Self-Organization-Map)聚类及BP(Back Propagation)建立换道意图识别模型,基于贝叶斯理论建立最小风险贝叶斯换道条件判别模型。模型开发和测试采用车辆轨迹数据集(NGSIM),提取数据中的换道行为特征参数作为模型的输入,将驾驶员换道决策预测视为输入变量的函数。通过对比最小贝叶斯和最小风险贝叶斯方法发现,由后者构建的换道条件判别模型效果较好,对于不换道行为的预测精度为90.4%,换道行为的预测精度为73.8%。鉴于错误的换道决策可能导致交通事故,而错误的不换道决策只会导致失去一次换道的机会,在换道辅助系统中,不换道决策的精确度要求需高于换道决策的精度。最后,在微观交通仿真系统中加入换道决策模型,其结果验证换道决策安全。最小风险贝叶斯换道条件判别模型的引入,使得换道决策系统能够通过修正风险系数,进一步提高换道判别精度,减少不安全的换道概率。

Safety driving lane change decision guide for the advanced drivers' assistance system

This paper would like to present a lane-changing decision model developed by its authors to provide some safety trafficlane changing guide for the drivers in hoping to reduce as much as possible the traffic accidents.As is well known,it is necessary to divide the lane-changing decisions into intention recognition and the traffic condition judgment so as to improve the prediction accuracy and build up the corresponding lane-changing model.However,for the above purpose,we have done the so-called SOM (Self-Organization-Map) cluster analysis and use the Neural network system in accordance with the results of the BP(Back Propagation) artificial neutral network lane-change intention recognition based on the application of the Bayes methods to the minimum error and the Bayes risk condition judgment model.What is more,the paper has also developed a model for examining the detailed vehicle trajectory data by using the Next Generation Simulation (NGSIM) data sets in addition to taking the characteristic parameters for the lane-change features as the inputs of the model.And thus as a result of the corresponding technical reconstruction,we have prepared a model for predicting the driver's decision on whether to change or not as a function.Comparing the lanechange decision model with the operating results of the Bayes minimum error and the Bayes minimum risk lets us believe that the latter turns to be better.And,finally,we have worked out the minimum Bayes risk method prediction accuracy which can be made to reach an accuracy rate of 90.4％ for the non-change events and 73.8％ for the change events.However,whereas a wrong decision to take a non-change event as a change event may result in a traffic crash,a wrong decision to take a change event as a non-change event can only result in an opportunity loss for the lane-change.Therefore,in a lane-change assistance system,it would be more critical to increase the prediction accuracy for the non-change events than for the change ones.And,so,integrating the decision model into the microscopic traffic simulation system,it would be possible to verify the simulation results of the safety of the lane-changing decision.Therefore,it can be said thatthe establishment of the minimum Bayes risk model can help to improve the analysis accuracy by correcting the risk parameters and reducing the unsafe changing lane probabilities in the same way as that the condition judgment model does.It has made the decision model better by sending a warning signal for the vehicle with the lane-change intention in spite of meeting no the change condition.