基于HMM的车辆行驶状态实时判别方法研究  

Determination of Real-time Vehicle Driving Status Using HMM

作　　者：王相海[1,2] 丛志环[1] 方玲玲[1,2] 秦钜鳌 

机构地区：[1]辽宁师范大学计算机与信息技术学院大连116029 [2]苏州大学江苏省计算机信息处理技术重点实验室苏州215006

出　　处：《自动化学报》2013年第12期2131-2142,共12页Acta Automatica Sinica

基　　金：国家自然科学基金(41271422);高等学校博士学科点专项科研基金(20132136110002);辽宁省自然科学基金(20102123);计算机软件新技术国家重点实验室开放基金(KFKT2011B09,KFKT2011B11);南京邮电大学图像处理与图像通信江苏省重点实验室开放基金(LBEK20 10003);智能计算与信息处理教育部重点实验室(湘潭大学)开放课题(2011ICIP06)资助~~

摘　　要：对交通视频车辆轨迹时序特征下的车辆行驶状态进行研究,提出了一种基于隐马尔科夫模型(Hidden Markov model,HMM)的车辆行驶状态实时判别方法.首先对轨迹序列进行了基于轨迹长度的去不完整轨迹序列、对车辆轨迹点序列的线性平滑滤波和最小二乘线性拟合的预处理操作,保证了所获得轨迹序列的有效性;其次,提出一种基于车辆运行轨迹点序列方向角的车辆轨迹特征值表示方法和基于方向角区间划分的HMM观察值序列生成方法,该方法以方向角的区间变化来区分不同轨迹模式的特征;最后,采用多观察值序列下的Baum-Welch算法训练得到相关交通场景轨迹模式类的最优HMM参数,并通过实时获取车辆行驶轨迹段与相应模型的匹配,实现对车辆行驶状态的实时判别.仿真实验验证了本文方法的有效性和稳定性.In this paper, we propose a method for determination of vehicle driving status from its time-ordered trajectory data using the hidden Markov model （HMM）. Firstly, we take some preprocessing including linear smooth filtering and least square fitting to abandon the trajectory sequences whose lengths are not enough, so as to guarantee the usability of acquired trajectory sequences. Secondly, we extract trajectory direction angle features from the trajectory sequences, and on this basis we propose a direction angle region partition algorithm to generate the observation sequences, which will determine the different trajectory patterns acquired by vehicle real-time various driving status. Finally, we get the optimal HMM model parameters of each trajectory pattern in specific traffic scene by multiple observations based Baum-Welch algorithm, then through matching with the above trained HMM models, we can determine the real-time vehicle driving status. Experiment results demonstrate the effectiveness and stability of this method.

关 键 词：视频车辆轨迹 隐马尔科夫模型 方向角 行驶状态 实时判别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U495[自动化与计算机技术—计算机科学与技术]