基于支持向量机的车辆驾驶行为识别研究

从车辆行驶轨迹的角度,车辆驾驶行为可细分为车辆跟驰行为、车辆换道准备行为和车辆换道执行行为,它们对交通拥堵、交通事故等都有着重要影响,也是自动驾驶、交通仿真等系统的基础构成模块.然而,如何从实际微观交通流数据中对3种行为进行识别是驾驶行为研究的基础和难点.本文提出基于支持向量机的驾驶行为识别方法,使用真实车辆轨迹数据,为提高模型的准确率,首先对样本数据进行归一化和主成分分析预处理,然后采用网格搜索算法对惩罚因子和核参数进行寻优,最后利用样本数据对基于支持向量机的分类模型进行训练和测试.结果表明,模型的测试精度达到了98.41%,能够很好地识别车辆的行驶状态,为驾驶行为各阶段的研究提供支持.

Driver Behavior Recognition Based on Support Vector Machine

From the viewpoint of the vehicle trajectory, driver behavior can be subdivided into the carfollowing behavior, lane- changing preparation behavior and lane- changing execution behavior. The three behaviors have a significant impact on traffic congestion and traffic safety, and are also the elemental modules of the automatic driving system and traffic simulation system. However, how to identify the three kinds of behavior from the real microscopic traffic flow data is a still unsolved problem. We propose a method of driver behavior recognition based on Support Vector Machine (SVM). The real vehicle trajectory data is used, which is normalized first and then processed using Principal Component Analysis. The grid search algorithm is adopted to find the optimal penalty factors and kernel parameters, and then the sample data is applied to train and test the proposed model. The results display that the accuracy of the model is 98.41%, which can identify the driver behavior with high performance and provide support for the research of driver behavior.