人车交互的行人过街方位概率动态预测模型

Dynamic Prediction Model of Pedestrian Crossing Orientation Probability Based on Human-vehicle Interaction

行人的安全是道路交通的最重要指标之一,由于行人行走轨迹多变,增加了自动驾驶汽车主动避障和轨迹规划的难度,精准高效的行人轨迹预测方法将有助于提高驾驶辅助系统效率和避障成功率,从而提高行人和车辆的安全性。为此,提出一种人车交互的行人过街方位概率动态预测模型,实现考虑人与车之间相互作用下的行人过街方位动态高效精准预测。首先,基于车辆对过街行人的影响,建立人车交互风险场模型,确定车辆尺寸、速度以及与行人的距离等参数对风险场的影响;其次,构建行人与车辆距离、行人与目标点距离以及行人过街步数的行人过街效益函数,并提出基于巢式概率模型的行人动态过街方位概率模型方法;然后,分析不同车辆类型、速度等特征对行人过街参数的影响规律;最后,通过试验采集分析某路段行人过街与车辆行驶轨迹数据,分析行人过街的规律。经试验与仿真结果对比,表明提出的方法能准确高效对行人过街方位进行预测,相较于长短期记忆网络(Long short-term memory,LSTM)的预测方法,计算效率提高了93.4%,精度提高了31.8%。

Pedestrian safety is one of the most important indicators of road traffic,due to the changeable trajectory of pedestrians,increasing the difficulty of active obstacle avoidance and trajectory planning of autonomous vehicles,accurate and efficient pedestrian trajectory prediction methods will help improve the efficiency of driver assistance systems and the success rate of obstacle avoidance,thereby improving the safety of pedestrians and vehicles.Therefore,this study proposes a dynamic prediction model of pedestrian crossing orientation probability based on human-vehicle interaction,which realizes efficient and accurate prediction of pedestrian crossing orientation dynamics considering the interaction between people and vehicles.Firstly,based on the influence of vehicles on pedestrians crossing the street,a human-vehicle interaction risk field model is established,and the influence of vehicle size,speed and distance from pedestrians on the risk field is determined.Secondly,the pedestrian crossing benefit function of pedestrian-vehicle distance,pedestrian-target point distance and pedestrian crossing steps is constructed,and a pedestrian dynamic crossing azimuth probability model method based on nested probability model is proposed.Then,the influence of different vehicle types,speeds and other characteristics on pedestrian crossing parameters is analyzed.Finally,the data of pedestrian crossing and vehicle trajectory in a certain road section are collected and analyzed through experiments,and the law of pedestrian crossing is analyzed.Comparing experimental and simulation results,it shows that the proposed method can accurately and efficiently predict the orientation of pedestrian crossings,and compared with the prediction method of long short-term memory(LSTM),the calculation efficiency is increased by 93.4%and the accuracy is improved by 31.8%.