基于HMM-SVM的驾驶员换道意图辨识研究  

Research on driver's lane change intention recognition based on HMM and SVM

作　　者：宋晓琳[1] 郑亚奇 曹昊天[1] 

机构地区：[1]汽车车身先进设计制造国家重点实验室,湖南大学长沙410082

出　　处：《电子测量与仪器学报》2016年第1期58-65,共8页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(51175159);国家自然科学基金(51575169);中华人民共和国人力资源和社会保障部留学人员科技活动项目资助

摘　　要：为了提高驾驶员换道意图的辨识率,提出了一种基于隐马尔可夫模型(HMM)和支持向量机(SVM)的混合模型。通过驾驶员在环仿真实验平台采集1.2 s时间窗内的驾驶员方向盘转角、油门踏板操作信息,匹配时序性良好的各个HMM模型(紧急左换道、正常左换道、紧急右换道、正常右换道和车道保持五种HMM模型)。然后结合各个HMM模型输出的最大似然估计值,由SVM进行分类,从而辨识出驾驶员当前的换道意图。仿真结果表明:相比单独的HMM或SVM,该混合模型能够更准确地辨识驾驶员的换道意图,辨识率高达98%,且耗时仅需0.006 s,具有较好的实时性。A hybrid model based on hidden Markov model （HMM） and support vector machine （SVM） is proposed to improve the recognition rate of the driver＇ s lane change intention. The driver＇ s steering wheel angle and accel- erator pedal data in 1.2 second time window are collected by the Driver-in-Loop （DiL） simulation experiments, these data could be matched with five HMM models （emergency left lane change, normal left lane change, emer- gency right lane change, normal right lane change and lane keep these five HMM models）, which possess an out- standing characteristic of time sequence. SVM can classify the maximum likelihood estimation which is outputted by HMM models. Finally, it can recognize the driver＇ s lane change intention. The simulation results show that this proposed hybrid model can recognize the driver＇ s lane change intention more accurately when compared with the classified approach only with the HMM or SVM, the recognition rate is reached as high as 98% , and takes only 0. 006 second, which shows that it has an excellent performance in real time.

关 键 词：驾驶员换道意图 混合模型 隐马尔可夫模型 支持向量机 

分 类 号：TN709[电子电信—电路与系统]