Nexar's real-world dashcam collision dataset

Nexar's real-world dashcam collision dataset 是一个专注于自我车辆参与的碰撞事件的真实世界行车记录仪数据集。该数据集包含 1.5k 个真实的行车记录仪视频，其中司机经历了真实的碰撞和险些发生的碰撞。该数据集提供了三个关键优势：1) 仅关注自我车辆参与的碰撞事件，消除了无关的碰撞事件；2) 包括通过紧急操作避免碰撞的近碰撞事件；3) 来自 10 个人类注释者的标准化、共识性的警报时间注释，使时间评估一致。该数据集对开发有效的 ADAS 方法至关重要，因为它捕捉了真实驾驶场景的真正分布，包括合成数据集无法复制的罕见事件。

Nexar's real-world dashcam collision dataset is a real-world dataset focused on collision events involving the ego vehicle, sourced from authentic dashcam footage. It contains 1.5k genuine dashcam videos capturing actual collisions and near-collisions experienced by drivers. The dataset offers three core advantages: 1) It exclusively targets collision events involving the ego vehicle, eliminating irrelevant collision incidents; 2) It includes near-collision events where collisions were averted via emergency driving maneuvers; 3) It provides standardized, consensus-based alert timing annotations from 10 human annotators, enabling consistent temporal evaluation. This dataset is critical for developing effective ADAS methodologies, as it captures the true distribution of real-world driving scenarios, including rare events that cannot be replicated by synthetic datasets.