多车行驶轨迹数据集

车辆行驶轨迹数据包括以受控车辆为中心，周围其他车辆的轨迹坐标文件，受控车辆被假设采取随机但安全的动作，跟踪以受控车辆为中心，其他车辆的轨迹变化。轨迹长度为30秒，保存的特征包括车辆的坐标和绝对速度。预处理之后的文件以受控车辆为中心，每个时刻所有其他车辆的特征被描述为车道网格中的点特征，是网格的轨迹。预处理之后的文件以受控车辆为中心，每个时刻所有其他车辆的特征被描述为车道网格中的点特征，是网格的轨迹。具体的预处理方式是（1）以受控车辆为中心，将每个时刻的车辆行驶轨迹提取出来，填入以受控车辆为中心的网格方阵；（2）将不同时刻的以车辆为中心的网格仿真串联成网格轨迹；（3）以递进的方式收集轨迹长度为3的轨迹片段，把下一时刻的网格作为预测网络输出的目标，以递进的形式处理所有的轨迹。最终形成用于深度学习的输入-输出对的形式。Vehicle driving trajectory data consists of trajectory coordinate files centered on the controlled vehicle, which contain trajectory information of surrounding other vehicles. The controlled vehicle is assumed to adopt random yet safe actions, and tracks the trajectory changes of other vehicles centered on itself. The trajectory duration is 30 seconds, and the stored features include vehicle coordinates and absolute speed. After preprocessing, the files are centered on the controlled vehicle, where the features of all other vehicles at each moment are represented as point features within the lane grid, forming grid-based trajectories. The specific preprocessing steps are as follows: (1) Extract the driving trajectories of vehicles at each moment centered on the controlled vehicle, and fill them into a square grid centered on the controlled vehicle; (2) Concatenate the vehicle-centered grid simulations across different moments into grid trajectories; (3) Collect trajectory segments with a length of 3 in a progressive manner, take the grid at the next moment as the target output of the prediction network, and process all trajectories in a progressive fashion. Finally, the dataset is structured into input-output pairs suitable for deep learning.

搜集汇总

背景与挑战

背景概述

该数据集由东南大学创建，包含以受控车辆为中心、周围其他车辆在高速公路环境下的行驶轨迹数据，轨迹长度为30秒，涵盖坐标和绝对速度特征。数据经过预处理，将车辆特征描述为车道网格中的点，形成用于深度学习的网格轨迹输入-输出对，总数据量为199.04MB，包含6002个文件。