基于多隐马尔可夫模型的车辆机动行为识别与预测  

Maneuver Behavior Identification and Prediction for Vehicles Based on Multi-hidden Markov Model

作　　者：曹凯[1,2] 于善义 于少伟[2] 

机构地区：[1]山东理工大学交通与车辆工程学院,山东淄博255049 [2]山东英才学院,山东济南250104

出　　处：《信息与控制》2014年第4期506-512,共7页Information and Control

基　　金：国家自然科学基金资助项目(61074140);山东省自然科学基金资助项目(ZR2010FM007)

摘　　要：针对车辆机动行为识别建模中多目标时空依赖关系的不确定性难以表述和刻画的问题,提出了基于多隐马尔可夫模型(M-HMM)的建模方法,对多种因素影响下的道路交通微观态势的时空相关性进行建模,识别和预测交通场景中车辆的机动行为.为了实现对车辆的机动行为识别,使用Baum-Welch算法和前向算法,生成了模型训练和模型评价两类输入数据;采用机动驾驶模拟方法,建立车辆机动行为数据库,提高了HMM的参数学习效率.高速公路的超车实验结果表明,在前方被观察车辆左轮压在车道线这一时刻,车道变换机动预测正确率已达到98.3%,而此前的0.4 s内,正确率约为75%.The uncertainty of multi-ohjective temporal dependencies is difficult to be expressed or described in the modeling for recognizing the maneuverability of vehicles. To solve this problem, we propose a modeling meth- od based on the multi-hidden Markov model （M-HMM）. We model the spatio-temporal dependencies of traf- fic micro-situations in the influence of many factors by the method to identify and forecast the maneuvering be- havior of vehicles in the traffic scenes. We apply the Baum-Welch algorithm and the forward algorithm to gen- erate two types of input data for model training and evaluation. We also build a database of vehicles＇ maneu- vering behaviors using a driving simulation method to improve the learning efficiency of the parameters of HMM. The experimental results of overtaking on the expressway show that at the point of time when the left tires of a vehicle passing lane the markers. The model＇s prediction accuracy of lane change is 98.3%, and the accuracy during the 0.4 s before true lane change is about 75%.

关 键 词：交通场景理解 隐马尔可夫模型(HMM) 机动行为识别 预测 

分 类 号：U471.15[机械工程—车辆工程] X913.3[交通运输工程—载运工具运用工程]