基于车辆视角数据的行人轨迹预测与风险等级评定  

Pedestrian Trajectory Prediction and Risk Grade Assessment Based on Vehicle-Perspective Pedestrian Data

作　　者：张哲雨 吕超 李景行 熊光明[1] 吴绍斌[1] 龚建伟[1] Zhang Zheyu;LüChao;Li Jinghang;Xiong Guangming;Wu Shaobin;Gong Jianwei(School of Mechanical Engineering,Beijing Institute of Engineering,Beijing 100081)

机构地区：[1]北京理工大学机械与车辆学院,北京100081

出　　处：《汽车工程》2022年第5期675-683,共9页Automotive Engineering

基　　金：北京理工大学科技创新计划前沿交叉与学科创新专项计划和国家青年自然科学基金(61703041)资助。

摘　　要：常用的基于路基视角数据的行人轨迹和风险预测模型往往无法避免复杂的建模运算和人工判断。为简洁而有效地预测行人轨迹和评定风险等级,本文中采用车辆视角数据建立行人轨迹和风险等级的预测模型,并先后进行车辆视角行人数据的采集、基于长短期记忆神经网络的行人轨迹预测和基于聚类分析-支持向量机方法的风险等级评定。实验结果表明,基于车辆视角数据所建立的数据驱动的模型能捕捉行人与车辆的运动趋势和交互特征,具有预测行人轨迹和评定风险等级的能力。The commonly used pedestrian trajectory and risk prediction model based on roadbed-perspec⁃tive data often cannot avoid complex modeling calculation and manual judgment.For succinctly and effectively pre⁃dicting pedestrian trajectory and evaluating risk grade,a pedestrian trajectory and risk grade prediction model is created based on vehicle-perspective pedestrian data in this paper.The acquisition of vehicle-perspective pedestrian data,the prediction of pedestrian trajectory based on long-short term memory neural network and the assessment of risk grade based on clustering analysis-support vector machine method are successively conducted.The results of experiments show that the data-driven model built based on vehicle-perspective pedestrian data can capture the movement tendency and interaction characteristics of pedestrian and vehicle and is capable of predicting pedestrian trajectory and assessing risk grade.

关 键 词：行人轨迹预测 行人风险等级评定 车辆视角行人数据 长短期记忆神经网络 聚类分析 支持向量机 

分 类 号：U463.6[机械工程—车辆工程]