基于轨迹数据的非机动车道内冲突事件自动识别与可视化

准确识别非机动车道内的交通冲突事件既是量化评价非机动车道安全水平的基础，也是剖析与理解非机动车道内运行风险产生与发展过程的重要前提。基于轨迹的交通冲突识别与分析技术已被运用于城市交叉口处机动车之间以及机非之间的交互安全分析，但目前鲜有聚焦非机动车道场景内交通冲突事件的分析方法。针对该问题，提出一种基于个体的（Agent-based）的非机动车道冲突风险自动识别方法。该方法以非机动车道内运动个体的轨迹时序数据为输入，将每条轨迹视为单独的运动个体。在每一时间步下，从个体视角切入，利用多坐标系转换思想进行个体和其视觉域内运动个体的追尾冲突、角度冲突与正面碰撞冲突判断。在此基础上，构建可对单一时间步下非机动车道内所有运动个体同时处理的冲突识别并行计算框架，能有效保证多个体场景下的冲突计算效率，在自动判别各类冲突风险的基础上，亦可对冲突事件信息进行提取与存储。基于该方法，可从宏观与微观层面以冲突事件统计、冲突热力分布图、冲突事件网络图与冲突个体微观行为图等形式对非机动车道内的运行风险进行可视化输出。通过案例分析，验证所提方法在非机动车道场景进行冲突风险识别的有效性，并结合具体输出结果探讨该方法应用的可能性。研究成果为非机动车道的运行风险评价提供了有效的量化手段，对于深入理解非机动车道内的冲突形成机理及风险场景中非机动车的微观行为模式具有积极意义。

Automated Identification and Visualization of Conflict Events in Bike Lanes Using Trajectory Data

Accurate identification of traffic conflicts is an essential precondition for quantifying safety performance and understanding the time-varying safety risks of bike lanes. Traffic conflict analytics based on trajectories have been widely applied in the safety analyses of vehicle-vehicle as well as vulnerable road user-vehicle interactions at urban intersections. However, there has been limited research on traffic conflict risks in bike lanes; therefore, an effective tool for identifying traffic conflicts between non-motorized vehicles is still lacking. In this regard, this study proposes an agent-based procedure for automatically identifying traffic conflicts in bicycle lanes. The developed procedure uses trajectory data as inputs and treats each trajectory as an individual agent. At each time step, different types of conflict risks between the agent and the objects within its view, including read-end, angle, and head-on conflicts, were examined via multiple coordinate transformations. Then, a parallel computing structure was established to improve the computation efficiency of conflict point estimations by parallelizing the conflict computation of all agents. Every time a conflict event was identified, the corresponding time stamp and conflict indicators were also recorded in the agent's motion sequence. The developed procedure provides a visualized demonstration of identified conflict events in both macroscopic and microscopic ways, such as conflict event statistics, heatmaps of conflict point distribution, conflict event graph network, and conflict-related time series diagram of involved agents. Subsequently, the effectiveness of the proposed procedure is demonstrated using real-world drone-captured trajectory data, and potential applications of the proposed method are discussed based on the obtained data. The results would be helpful in the analysis and evaluation of operational risks in bicycle lanes, and could provide insights into understanding the traffic conflict mechanism in bike lanes as well as microscopic behavioral patterns of non-motorized vehicles under risky scenarios.