The continuous movement process of vehicle should be processed by state discretization when the recognition of driving behavior and the prediction of vehicle state are performed.During state division,it would esaily appear the repeated state divisions and it is difficult for a hidden markov model (HMM) to determine state transition probability and the initial probability during identifying state sequences.Therefore,this paper adopts clustering analysis to process the datas in the sliding window so as to ensure the uniqueness of state division.In addition,the state logical decision is established to correct the wrong state division.The different driving behaviors are divided into several layers through CPNtool hierarchical analysis software.The interaction between layers is achieved through the state transfer between layer and layer.The time which state flows in the constant intervals occupy is used to the current driving behavior.Finally,the sample datas collected from the test car are used to validate the established model and the result shows that this model can show the switch of different driving behavior states in visual way.The model can reached more than 96％ judgment accuracy for driving behavior.%在驾驶行为识别以及车辆运动状态预测时,需将车辆连续运动过程离散化成状态,对状态划分时极易出现状态重复划分的情况,且利用隐马尔科夫模型在对行为状态序列进行判定时难以准确确定各个状态之间的转移概率以及各个状态的初始概率,因此,提出利用聚类分析对滑动时窗内的数据进行聚类,确保驾驶过程中状态划分的唯一性,此外,利用模糊逻辑规则对出现的异常行为状态进行修正.利用CPNtool层次化分析软件,将不同驾驶行为分成若干层,通过状态之间的转移实现层与层之间的交互,通过判定等时间段内状态流所占用某个层的时间比例来确定当前的驾驶行为.最终利用测试车采集到的样本数据对所建模型进行有效性验证,结果表明,该模型能够以可视化的方式展现不同驾驶行为之间的状态切换,对当前的驾驶行为的判断准确率达到了96％以上.
展开▼
