Abstract: Pedestrians are an indispensable part of the traffic system, and they are also the most flexible, complex, and changeable factors in the traffic scene. According to statistics, more than 90% of traffic accidents are caused by human factors. This paper mainly focuses on pedestrians and studies the early warning method for pedestrians crossing the street. It is established a microscopic interaction model of pedestrians and vehicles, and proposed an attention degree algorithm based on the deviation angle of the pedestrian’s head; collected a bird’s-eye view video of pedestrians and vehicles, calibrated the two-dimensional coordinate points of pedestrians and vehicles as trajectory data, and using trajectory data and heads, deviation angle calculates the maximum acceleration and the degree of attention, and clusters the two typical crossing styles with the maximum acceleration and the degree of attention. Analyze the warning degree of different street crossing styles. Finally, the generative confrontation network model is used to predict the trajectory of pedestrians during human-vehicle interaction; trace with a predicted trajectory describe the reaction of pedestrians to interactive vehicles, this experiment is useful for studying unsignalized intersections. The movement style and trajectory distribution of pedestrians crossing the street have important reference significance. Theoretically, the safety of pedestrian vehicle interaction at unsignaled intersections is enhanced.

Key words: Pedestrian crossing style; Human-vehicle micro-interaction model; Attention algorithm; Trajectory prediction; Warning for pedestrians crossing the street

摘要： 行人是交通系统中不可或缺的一部分，也是交通场景内最灵活、复杂、多变的因素。 据统计，90%以上的交通事故都是由人为因素造成的。文章主要面向行人，研究行人过街的 预警方法。建立人车微观交互模型，提出以行人头部偏差角为基础的关注度算法。采集行人 和车辆鸟瞰视频，标定行人与车辆的二维坐标点为轨迹数据，通过轨迹数据和头部偏差角计 算出最大加速度和关注度，以最大加速度和关注度聚类出两种典型过街风格，分析不同过街 风格的预警程度。最后利用生成对抗网络模型预测出人车交互时行人的轨迹，用预测轨迹描 述行人对交互车的反应情况。该实验对研究无信号十字交叉路口行人过街预警有重要参考意 义，理论上增强了行人与车辆在无信号交叉路口的交互安全性。

关键词: 行人过街风格；人车微观交互模型；关注度算法；轨迹预测；行人过街预警