协同自适应巡航控制环境下车辆轨迹数据重构方法

A Method for Vehicle Trajectory Data Reconstruction in Cooperative Adaptive Cruise Control Environment

针对国内尚缺乏协同自适应巡航控制环境下的车辆轨迹数据集,且现有轨迹重构方法存在处理单一、缺乏实测数据验证等问题,设计了协同自适应巡航控制环境下交通流视频数据采集试验并组织实施。所采集的视频数据包含调研路段所有非自动驾驶车辆行驶信息以及3辆自动驾驶车辆的行驶信息,基于视频识别技术对所采集视频进行提取,获得了原始车辆轨迹数据。针对原始车辆轨迹数据深度分析了误差来源,融合拉格朗日算法与卡尔曼滤波理论,设计了两步轨迹数据重构方法。识别速度和加速度异常值并反向修正了车辆位置数据,直至速度和加速度都在正常范围之内,实现了对数据集异常数据的插值替换以及降噪处理。最后,基于车辆实测数据对重构后的轨迹数据进行了验证,以避免出现降噪不足或过拟合现象。结果表明:两步轨迹数据重构方法能够有效修正原始车辆轨迹数据并保留原始数据的结构特征;通过对插值效果进行分析发现,受原始数据时间粒度影响,拉格朗日3次插值比5次插值收敛速度更快,因此,应结合数据特征与计算工作量确定拉格朗日次数;通过对卡尔曼滤波效果进行分析发现,观测噪声协方差与系统噪声协方差比值对重构误差产生影响,该比值的确定应满足误差要求,且避免出现车辆轨迹数据重构过拟合现象。

In view of the lack of domestic vehicle trajectory data set in cooperative adaptive cruise control environment,and the existing trajectory reconstruction methods have problems of single processing and lack of actual measurement data verification,etc,the traffic flow video data collection test in cooperative adaptive cruise control environment is designed and implemented.The collected video data contain the driving information of all non-autonomous vehicles on the investigated road section as well as the driving information of 3 autonomous vehicles,the collected videos are extracted based on video recognition technology,and the original vehicle trajectory data are obtained.In view of the original vehicle trajectory data,the error sources are analyzed in depth,and a 2-step trajectory data reconstruction method is designed by integrating Lagrange algorithm and Kalman filtering theory.The abnormal values of speed and acceleration are identified,and the vehicle position data are corrected reversely until the values of speed and acceleration are all within the normal range to realize the interpolation replacement of abnormal data in the data set and the denoising.Finally,the reconstructed trajectory data are verified based on the vehicle measured data to avoid the phenomenon of insufficient denoising or overfitting.The result shows that(1)the original vehicle trajectory data can be corrected effectively and the structural characteristics of the original data can be retained by using the 2-step reconstruction method;(2)it is found through the analysis of the interpolation effect that influenced by time granularity of original data,the convergence speed of cubic Lagrange interpolation is higher than those of quintic Lagrange interpolation,therefore,the degree of using Lagrange algorithm should be determined by combining data characteristics and computational workload;(3)it is found through the analysis of Kalman filtering effect that the reconstruction error is affected by the ratio of observed noise covariance to systematic noise covariance,and the ratio should meet the requirements on errors and avoid overfitting of vehicle trajectory data reconstruction.