Zhang Howard, Liu Ash, Habibi Saied, Mohrenschildt Martin V, Ahmed Ryan
Department of Computing and Software, McMaster University 1280 Main St. W, Hamilton, ON, L8S 4L8, Canada.
Center for Mechatronics and Hybrid Technologies (CMHT), Department of Mechanical Engineering, 200 Longwood Road South, Unit 224, Hamilton, ON, L8P0A6, Canada.
Data Brief. 2025 Apr 11;60:111552. doi: 10.1016/j.dib.2025.111552. eCollection 2025 Jun.
Standardized datasets are essential for the development and evaluation of autonomous driving algorithms. As the types of sensors available to researchers increase, datasets containing a variety of temporally and spatially aligned sensors have become increasingly valuable. This paper presents a driving dataset recorded using a complete sensor suite for research on autonomous driving, perception, and sensor fusion. The dataset consists of over 9000 frames of data recorded at 10-20Hz using a complete sensor suite made up of Velodyne LiDAR, GPS/IMU, mm-wave radar, as well as color and infrared cameras. The capture scenarios include poor weather/lighting conditions, such as rain/night scenarios, and diverse traffic conditions, such as highways and cities with various objects. Both fully synchronized data and raw recordings in the form of ROS2 bags are provided, as well as 3D tracklet labels for individual objects. This paper provides technical details on the driving platform, data format, and utilities.
标准化数据集对于自动驾驶算法的开发和评估至关重要。随着研究人员可用传感器类型的增加，包含各种时间和空间对齐传感器的数据集变得越来越有价值。本文展示了一个使用完整传感器套件记录的驾驶数据集，用于自动驾驶、感知和传感器融合研究。该数据集由使用由Velodyne激光雷达、GPS/IMU、毫米波雷达以及彩色和红外摄像头组成的完整传感器套件以10 - 20Hz记录的超过9000帧数据组成。捕获场景包括恶劣天气/光照条件，如下雨/夜间场景，以及各种交通状况，如高速公路和有各种物体的城市。提供了完全同步的数据和ROS2包形式的原始记录，以及单个物体的3D轨迹标签。本文提供了关于驾驶平台、数据格式和实用工具的技术细节。
