面向驾驶员辅助系统的换道意图辨识方法研究

Method for recognizing the driver's lanechanging intention available for the advanced driver's assistant system

为提高先进驾驶员辅助系统工作效能,通过差异性分析选取了人-车-路特征参数,建立了支持向量机换道意图辨识模型。基于驾驶模拟器采集的数据,运用ROC曲线对比分析了不同参数组合、时窗、数据表达形式下模型的分类效果,结果表明以方向盘转角、加速度、车辆距车道线距离和驾驶人头动信息为特征参数,时窗为1.5s,以时窗内信息均值和方差表述特征参量可获得最优辨识效果。研究表明,模型在保证1s前瞻性的同时辨识率可达93%。

In this paper,we are attempting to introduce a method for efficient recognition of the driver＇ s lane changing intention.As is known,it is of great importance to identify the driver＇s intention for lane change in the fast driving process and improve their performance by learning from the advanced drivers＇ skills in controlling the vehicles.For this purpose,we have first of all made different analyses based on the data acquired from the driving simulator in their driving tasks.In so doing,we have found a series of influential factors,such as the steering direction,the vehicle acceleration,the driving space from other vehicles to the lane and the ahead-driving movement information,which are supposed to play the dominant role in lane keeping and lane change intention identification for the drivers to find the best supplies from the SVM model.For this purpose,we have also chosen four different feature-variable combinations,three different time windows and two different data representation methods.According to the aforementioned variables,we have also laid out different model feature-sets.And,thirdly,we have managed to load out 24 feature-sets to the SVM model for the model-control training,and a receiver operating characteristic（ROC） curve for assessing the recognition skills in the above-mentioned training practice.And,consequently,we have set up the area under the ROC convex hull and the AUC for comparing the model classification behaviors.The results of our study demonstrate that the SVM model can function successfully in recognizing the objectives on the condition when the feature-set works by steering the angle,the vehicle accelerator,the distance from the vehicle to the lane,as well as the driver＇s heading movement information.In addition,we have also determined the time window of 1.5seconds,and the mean and variance of inputs for the data representation.What s more,the cross-validation has been used to optimize the model parameters.And,finally,we have offered an experiment verification to test the optimized-model,the successful application of the trial-application results,which prove that the model is capable to predict the lane changes in one second in advance with an accuracy as precise as to 93%.