Real-world naturalistic near-miss driving dataset

该数据集由东京农业技术大学提供的行车记录仪视频组成，包含了超过22万个视频片段，这些片段记录了接近事故的 safety-critical 事件，并附有详细描述标签。每个视频大约15秒长，捕捉了关键事件前10秒以及随后的5秒内容。视频均配有风险分析和事件描述信息。该数据集规模宏大，包含22万个视频片段，其中1000个样本用于评估。任务集中在复杂风险场景下的规避决策制定。

This dataset consists of dashcam videos provided by Tokyo University of Agriculture and Technology, containing over 220,000 video clips that record near-accident safety-critical incidents with detailed descriptive labels. Each video is approximately 15 seconds long, capturing 10 seconds prior to the critical incident and the subsequent 5 seconds following the event. All videos are accompanied by risk analysis and incident descriptions. This large-scale dataset reserves 1,000 samples for evaluation. The tasks focus on avoidance decision-making in complex risk scenarios.