A3D 交通事故数据集

A3D 全称 AnAn Accident Detection，是一个全新的关于各种交通事故的数据集。识别交通违章和自动驾驶场景中的事故事件对于自动驾驶和驾驶员辅助系统至关重要。常见的交通事故视频往往假设摄像头固定，视频有静态背景，但这对于车载摄像头就不合理了。所以该数据集的视频基于一种无监督的第一人称视频交通事故检测方法，主要通过预测交通参与者的未来位置检测异常。该数据集包含来自 YouTube 的视频和一个 .pkl 文件，包括人工标注的视频开始/结束时间和异常开始/结束时间。

A3D, fully named AnAn Accident Detection, is a novel dataset focused on a wide range of traffic accidents. Identifying traffic violations and accident events in autonomous driving scenarios is of critical significance for both autonomous driving systems and driver assistance systems. Most existing traffic accident video datasets assume fixed cameras and static backgrounds, a setting that is impractical for in-vehicle camera setups. Accordingly, the videos in this dataset are based on an unsupervised first-person traffic accident detection approach, which detects anomalies primarily by predicting the future positions of traffic participants. This dataset includes videos sourced from YouTube and a .pkl file containing manually annotated start and end timestamps for both the videos themselves and the anomalous accident events.

搜集汇总

背景与挑战

背景概述

A3D 交通事故数据集是一个专注于自动驾驶场景中事故检测的数据集，其特点在于采用无监督的第一人称视频方法，通过预测交通参与者未来位置来识别异常，从而弥补了传统固定摄像头假设的不足。数据集包含YouTube视频和.pkl文件，并提供了人工标注的视频及异常时间范围，适用于自动驾驶和驾驶员辅助系统的开发与评估。