基于毫米波雷达组群的全域车辆轨迹检测技术方法

高精度车辆轨迹数据对于高速公路交通管理和智慧服务具有非常重要的研究及应用价值，然而现有的车辆轨迹感知技术难以获得全域全时车辆轨迹数据。为此，提出一种基于毫米波雷达的全域车辆轨迹跟踪技术方法，该方法包括：雷达原始数据获取及适配、轨迹数据清洗及降噪、道路线形感知及还原、车辆轨迹匹配及拼接。其中，雷达原始数据获取及适配通过构建雷达帧数据适配表将雷达数据格式标准化，并通过构建的轨迹可信度评价指标K，剔除镜像车辆轨迹数据，进而基于历史行车轨迹的统计学特征，采用聚类方法还原道路线形，最终通过雷达群组间车辆轨迹特征分析及匹配拼接，实现设备内部及跨设备对车辆轨迹的持续跟踪。利用载波相位差分技术(Real-time Kinematic, RTK)和基于无人机航拍视频定位技术分别对单车及多车轨迹跟踪精度进行检验。研究结果表明：在单目标跟踪状态下，系统的纬度偏差均值为-0.284 m，经度偏差均值为-0.352 m，纬度误差均值为0.712 m，经度误差均值为0.539 m；在多目标跟踪状态下，系统丢车率约为8%，轨迹定位与真实位置偏差均值为0.990 m，具备良好的轨迹跟踪精度。该方法为未来从更加宏观的范围内研究个体驾驶行为风险转移分析、微观水平的驾驶风险的时空演化提供了数据支撑。

Road-range Tracking of Vehicle Trajectories Based on Millimeter-wave Radar

Accurate traffic flow data, particularly vehicle trajectory data with rich spatial-temporal information, have very important research and application value for expressway traffic management and intelligent services. However, owing to the limitation of a single sensor, it is difficult to obtain the road-range level trajectory data in real time using existing vehicle trajectory sensing technology. To this end, a vehicle trajectory perception method based on millimeter-wave radar (MMW) is proposed. The method includes the adaptation of the original MMW radar data, data cleansing and noise reduction, road alignment perception, trajectory matching, and splicing. The adaptation method standardizes the radar data format by constructing an adaptation table. Mirrored trajectory data were eliminated based on the reliability evaluation index K. A road alignment perception method was proposed based on the statistical characteristics of historical trajectories. To overcome the limitations of the perception range, trajectory matching and splicing methods were used to obtain a continuous trajectory within and across devices. Real-time kinematic and video data from an unmanned aerial vehicle were used to test the perception accuracy of the single-vehicle tracking and multiple object tracking of MMW radar systems. The verification results show that in the single-vehicle tracking state, the mean latitude offset was -0.284 m, the mean longitude offset was -0.352 m, the average error in latitude error was 0.712 m, and the average error in longitude error was 0.539 m. In the multiple-object tracking state, the false negative rate of the system was approximately 8%, and the average offset between the track positioning and real position was 0.990 m, which satisfies the accuracy requirements. This study provides data support for future studies on the risk transfer analysis of individual driving behaviors, spatial-temporal evolution of driving risks at the micro level, impact of traffic accidents, and migration mechanism of traffic congestion in a more macro scope.