基于换道行为多参数融合预测的安全度辨识方法研究

The existing lane-changing assistance system regards turn signal as the main basis to predict drivers’ lane-changing behavior, if there exists target vehicle within the given distance, the system will issue warning signals to the drivers. In real application, owing to the missing and false alarms have not been effectively solved, the system could not achieve its desired assistance driving function. Based on the conducted real-world experiments, dynamically collect the multi-source parameters, the research focus on the following aspects: Firstly, explore the change rules of drivers’ visual search mode, vehicle motion states and driving environment parameters in the intent time window, lane change mapping parameters are extracted, then mapping mechanism and time series discrepancies of different characteristic parameters on lane changing behavior are revealed in turn. In addition, consider the influence of road alignment, the positional relation between ego vehicle and surrounding vehicles is modeled to lock the conflict targets. Furthermore, an improved support vector machine algorithm, SFLA-SVM is designed to build the integrated fusion-prediction model for lane changing behavior. Finally, constructing the mapping parameters collection of lane changing safety, for the goal of online warning, LEP algorithm and evidence theory are applied to build the safety identification model of drivers’ lane changing behavior. The research results may enrich and perfect the prediction theory of driving behaviors, moreover, provide theoretical and technical support for the improvement of intelligent vehicles' active safety technologies, as well as the performance optimization of lane change warning system.

现有辅助系统以转向灯信号预测驾驶人的换道行为，当给定距离范围内识别出目标车辆，即对驾驶人进行安全预警。系统在实际工作过程中，漏警及虚警问题突出，无法实现预期的辅助驾驶功能。本项目拟组织真实道路环境下的实车试验，动态采集多源特征参数，开展以下研究：探索意图时窗内驾驶人视觉搜索模式、车辆运动状态以及行驶环境参数的变化规律，萃取驾驶人换道行为映射参数，揭示不同特征参数对换道行为的映射机理及时序差异；考虑道路线形的影响，建立自车与周围车辆的位置关系辨识模型，锁定冲突目标；基于蛙跳算法改进的支持向量机理论构建驾驶人换道行为一体化融合预测模型；构造换道行为安全度映射参数集合，应用LEP算法与证据理论建立面向于在线预警的驾驶人换道行为安全度辨识模型。研究成果将丰富、完善驾驶行为识别及预测理论，为智能车辆主动安全技术的改进、换道预警系统的性能提升提供重要的理论与技术支撑。

现有辅助系统以转向灯信号预测驾驶人的换道行为，当给定距离范围内识别出目标车辆，即对驾驶人进行安全预警。系统在实际工作过程中，漏警及虚警问题突出，无法实现预期的辅助驾驶功能。本项目拟组织真实道路环境下的实车试验，动态采集多源特征参数，开展以下研究：基于车辆运行状态及驾驶人操作特性，确定换道起始时刻，结合驾驶人视觉表征参数特征，确定驾驶人换道意图时窗。基于前景理论和驾驶期望满足度分析驾驶人换道决策机制。探索意图时窗内驾驶人视觉搜索模式、车辆运动状态以及行驶环境参数的变化规律，萃取驾驶人换道行为映射参数，揭示不同特征参数对换道行为的映射机理及时序差异；考虑道路线形的影响，建立自车与周围车辆的位置关系辨识模型，锁定冲突目标；基于GA-SVM算法建立换道行为一体化融合预测模型；构造换道行为安全度映射参数集合，应用LEP算法与证据理论建立面向于在线预警的驾驶人换道行为安全度辨识模型。研究成果将丰富、完善驾驶行为识别及预测理论，为智能车辆主动安全技术的改进、换道预警系统的性能提升提供重要的理论与技术支撑。