路侧感知车辆轨迹数据质量智能评估方法

智慧公路布设了大量路侧智能传感器，可以获取全时空车辆运行轨迹数据。然而，如何实现轨迹数据质量高效便捷的评估一直是困扰行业管理部门的难题。现有评估方法大多存在量化指标维度单一、鲁棒性较差等问题。为此，提出一种通过挖掘轨迹数据多维特征以快速评估轨迹数据质量的方法。首先基于轨迹多元信息从元素特征、时序特征和空间特征3个维度设计轨迹合理性、波动性与交互异常性评估指标，并分析评估指标与轨迹数据质量水平的相关关系；在此基础上提出一种利用多元评估指标实现轨迹质量评估的自适应融合回归模型；最后，结合公开轨迹数据集和实测数据集对指标和模型的可靠性及稳定性进行测试与验证。结果表明：轨迹合理性、波动性指标与数据质量显著相关，可基于此构建指标融合模型评估轨迹质量，且引入提出的交互异常性指标可较好地提升模型评估效果。随着模型得分的降低，轨迹数据的运动与交互特征的异常程度增大，持续时间增加。提出的智能评估模型可以挖掘评估指标与轨迹质量的关系，对不同质量水平的轨迹均保持较好的评估效果且优于传统的单一维度评估指标方法，具有良好的稳定性、鲁棒性和优越性，可为车路协同环境下海量的路侧感知轨迹数据提供可靠的质量评价与监测方法。

An Intelligent Quality Assessment Method for Vehicle Trajectory from Roadside Perception

Intelligent roadside sensors have been widely deployed in the latest leading smart highways to collect time and space vehicle trajectory data. However, efficiently evaluating data quality is a challenge for management departments. The methods used in previous studies evaluated data by adopting one-dimensional indicators and were not robust. In this paper, a method to rapidly assess the quality of vehicle trajectories by mining multiple characteristics has been proposed. First, three types of quality indicators were designed, that is, rationality, volatility, and interactive abnormality, and the correlations between the quality indicators and the quality of the trajectory were analyzed. Then, an adaptive fusion regression model based on the quality indicators was proposed to evaluate the quality of the trajectories. Finally, data obtained from the highD dataset and the real world were used to verify the effectiveness and robustness of the proposed model. Adding interactive abnormality indicators can improve the effect of model. The proposed model is better at distinguishing different qualities of trajectories. Compared with the traditional method that uses single-dimensional indicators, this model can accurately predict the quality of the trajectory data of different scores. In summary, as the model score decreases, the kinematic characteristics of the trajectory data become abnormal, the degree of interactive abnormality becomes higher, and the abnormality duration increases. The model is robust, stable, and superior, and it can provide reliable quality evaluation and monitoring methods for roadside perception trajectory data in cooperative vehicle infrastructure system (CVIS).