DeepScenario Open 3D Dataset (DSC3D)
收藏资源简介：
DeepScenario Open 3D数据集（DSC3D）是一个高质量的、无遮挡的、包含14种交通参与者的6自由度边界框轨迹数据集。该数据集包含超过17.5万条轨迹，数据规模和多样性显著超过了现有数据集，覆盖了复杂的车辆-行人交互和全面的停车操作等场景。数据集在五个不同地点录制，包括停车场、拥挤的市中心、陡峭的城市交叉口、联邦公路和郊区的交叉口。DSC3D数据集的目的是通过提供详细的3D环境表示来增强自动驾驶系统，从而提高障碍物交互和安全。数据集的应用领域包括运动预测、运动规划、场景挖掘和生成反应式交通代理等。数据集和相关工具在app.deepscenario.com上公开提供，以促进运动预测、行为建模和安全验证的研究。
The DeepScenario Open 3D Dataset (DSC3D) is a high-quality, occlusion-free trajectory dataset equipped with 6-degree-of-freedom (6DoF) bounding boxes covering 14 categories of traffic participants. This dataset contains over 175,000 trajectories, with its scale and diversity significantly exceeding those of existing datasets, and covers scenarios such as complex vehicle-pedestrian interactions and comprehensive parking maneuvers. The dataset was recorded at five distinct locations, including parking lots, crowded downtown areas, steep urban intersections, federal highways, and suburban intersections. The goal of the DSC3D dataset is to enhance autonomous driving systems by providing detailed 3D environmental representations, thereby improving obstacle interaction and driving safety. Application domains of the dataset include motion prediction, motion planning, scene mining, and the generation of reactive traffic agents, among others. The dataset and its associated tools are publicly available at app.deepscenario.com to facilitate research in motion prediction, behavior modeling, and safety validation.
DeepScenario Open 3D Dataset (DSC3D) 概述
数据集简介
- 名称：DeepScenario Open 3D Dataset (DSC3D)
- 类型：高精度、无遮挡的6自由度边界框轨迹数据集
- 采集方式：通过新型单目相机无人机跟踪流程获取
- 主要特点：
- 包含超过175,000条轨迹
- 涵盖14种交通参与者类型
- 提供详细的环境3D表示
数据集内容
- 数据量：约15小时的数据
- 采集地点：德国和美国的五个不同地点
- 停车场 (SIFI)
- 内城环境 (MUC)
- 非信号化交叉路口 (STR, SFO)
- 联邦高速公路 (BER)
- 轨迹数量：177,151条独特轨迹
分类框架
- 主要类别：
- 行人 (140,227)
- 自行车 (17,736)
- 汽车 (13,241)
- 滑板车 (1,475)
- 摩托车 (1,054)
- 动物 (677)
- 卡车 (475)
- 公共汽车 (191)
- 其他 (2,075)
应用场景
- 运动预测
- 运动规划
- 场景挖掘
- 生成反应式交通代理
获取方式
- 数据集及许可证可通过DeepScenario的web应用获取：app.deepscenario.com
引用格式
bibtex @inproceedings{dsc3d, title = {Highly Accurate and Diverse Traffic Data: The DeepScenario Open 3D Dataset}, author = {Dhaouadi,Oussema and Meier, Johannes and Wahl, Luca and Kaiser, Jacques and Scalerandi, Luca and Wandelburg, Nick and Zhuo, Zhuolun and Berinpanathan, Nijanthan and Banzhaf, Holger and Cremers, Daniel}, booktitle = {2025 IEEE Intelligent Vehicles Symposium}, year = {2025}, organization = {IEEE} }
- 1Highly Accurate and Diverse Traffic Data: The DeepScenario Open 3D Dataset慕尼黑工业大学（TU Munich） · 2025年
