| 引用本文: | 彭金栓,付锐,郭应时.自然驾驶条件下驾驶人换道行为实时预测[J].哈尔滨工业大学学报,2015,47(9):119.DOI:10.11918/j.issn.0367-6234.2015.09.022 |
| PENG Jinshuan,FU Rui,GUO Yingshi.Real-time prediction of lane-changing behaviors under naturalistic driving condtions[J].Journal of Harbin Institute of Technology,2015,47(9):119.DOI:10.11918/j.issn.0367-6234.2015.09.022 |
|
| 摘要: |
| 为降低车道变换的风险性,提出一种基于驾驶人视觉特性与车辆运动状态预测车道变换行为的方法. 应用视觉追踪系统、毫米波雷达等仪器设备,进行了真实环境下的实车驾驶试验. 基于换道前驾驶人后视镜注视特性确定换道意图时窗大小为5 s,构建换道行为预测的表征指标体系. 设计BP神经网络结构,构建换道行为预测模型. 结果表明：模型可以至少提前1.5 s预测驾驶人的换道行为,且预测精度达到95.58%. 与基于转向灯状态预测驾驶人换道行为相比,其预测精度及时序特性均有显著提升,证明了预测指标及预测方法的有效性. |
| 关键词: 驾驶行为 实车试验 车道变换 行为预测 BP神经网络 意图时窗 |
| DOI：10.11918/j.issn.0367-6234.2015.09.022 |
| 分类号:U471.15 |
| 基金项目:国家自然科学基金 (9,3)；教育部高等学校博士学科点专项基金(20135522110003)；中央高校基本科研业务费专项资金(2014G1502015). |
|
| Real-time prediction of lane-changing behaviors under naturalistic driving condtions |
|
PENG Jinshuan1,FU Rui2,GUO Yingshi2
|
|
(1.School of Transportation, Chongqing Jiaotong University, 400074 Chongqing, China; 2.School of Automobile, Chang’an University, 710064 Xi’an, China)
|
| Abstract: |
| To reduce the risk of lane changing behaviors, based upon drivers’ visual characteristics and vehicle motion states, a method for lane change prediction is proposed. By using visual tracking system, millimeter-wave radar and so on, the research group conducts experiments under real road environment. Based on drivers’ fixation characteristics of the rearview mirrors before lane change occurs, lane changing intent time window is determined as 5 s, the characteristic index for predict lane changing behavior is further built. By designing BP neural network, the lane change prediction model is constructed. Results show that the model may predict drivers’ lane changing behavior for at least 1.5 s in advance, and the prediction accuracy can reach 95.58%. As compared to predict lane change behavior via turn signals, the prediction accuracy and time series characteristics are both improved remarkably, thus verifying the effectiveness of the predictive index and method. |
| Key words: driving behavior real-world experiment lane change behavior prediction BP neural network intent time window |
