highD德国高速公路车辆轨迹数据集

highD数据集由德国亚琛工业大学汽车工程研究所发布，是基于无人机拍摄的德国高速公路的大型自然车辆轨迹数据交通视频，记录的数据包括来自六个地点的11.5小时测量值和110 000车辆（包括卡车和轿车），所测量的车辆总行驶里程为45 000 km，还包括了5600条完整的变道记录，数据本身不涉密。基于该数据集使用基于深度学习的视觉定位算法可对车辆定位误差小于十厘米，可用于车辆微观轨迹预测，并基于预测轨迹进行宏观交通流预测、冲突检测等。The highD dataset is published by the Institute of Automotive Engineering at RWTH Aachen University, Germany. It is a large-scale natural traffic video dataset with vehicle trajectory data captured by drones on German highways. The dataset includes 11.5 hours of measurement data and 110,000 vehicles (including trucks and passenger cars) from six locations, with the total driving mileage of all measured vehicles reaching 45,000 km. Additionally, it contains 5,600 complete lane change records, and the dataset itself is not classified. Using deep learning-based visual positioning algorithms on this dataset can achieve vehicle positioning errors of less than 10 centimeters. It can be used for microscopic vehicle trajectory prediction, as well as macroscopic traffic flow prediction and conflict detection based on the predicted trajectories.

搜集汇总

背景与挑战

背景概述

该数据集是一个德国高速公路车辆轨迹数据集，由德国亚琛工业大学基于无人机视频采集，包含11.5小时测量值、11万辆车和5,600条变道记录，总行驶里程达4.5万公里，数据不涉密。它主要用于自动驾驶领域的车辆微观轨迹预测和宏观交通流分析，支持交通冲突检测等研究，数据量约为837.72MB，由东南大学发布并隶属于国家重点研发计划项目。