The large-scale trajectory data provide the potential opportunity to a better understanding of driving behaviour for transportation applications and research. However, limited effort has been paid to the study on the evaluation of unsafe driving behaviour (UDB) based on trajectory data. In this work, the authors propose a four-layer processing framework for evaluation of driving behaviour using trajectory data, and first analyse the statistical distribution of various factors and mine UDBs from trajectory data by measuring the deviation from a normal distribution. Then, a membership function is designed to evaluate the severity rating of UDBs, and finally, an analytical hierarchy process-based method is employed to analyse UDBs both qualitatively and quantitatively. With experiments on trajectory data derived from August to September 2018 in Jiangxi, China, vehicles were classified in levels of risk, and the result shows that the proposed method offers a feasible and applicable way for transportation enterprises and drivers to monitor driving behaviour in real-time.

中文翻译：

---

分层定量分析，可从大量轨迹数据中评估不安全驾驶行为

大规模的轨迹数据为更好地了解运输应用和研究的驾驶行为提供了潜在的机会。但是，基于轨迹数据对不安全驾驶行为（UDB）评估的研究付出了有限的努力。在这项工作中，作者提出了一种使用轨迹数据评估驾驶行为的四层处理框架，并首先通过测量与正态分布的偏差来分析各种因素的统计分布并从轨迹数据中挖掘UDB。然后，设计了隶属度函数来评估UDB的严重性等级，最后，采用基于层次分析法的方法对UDB进行定性和定量分析。