CycleCrash
收藏资源简介:
CycleCrash是由加拿大皇后大学创建的一个新型数据集,专注于自行车碰撞视频的收集与分析。该数据集包含3000个行车记录仪视频,共计436,347帧,涵盖了从碰撞到安全互动的各种关键情境。数据集的创建过程包括从多个在线平台收集视频,并进行时间裁剪、文件转换、空间裁剪和缩放、帧率调整以及颜色归一化等预处理步骤。CycleCrash数据集的应用领域主要集中在自动驾驶和驾驶辅助研究社区,旨在通过提供详细的碰撞相关、骑车者相关和场景相关的标注,解决自行车安全问题,如碰撞预测和时间到碰撞的预测。
CycleCrash is a novel dataset developed by Queen's University in Canada, dedicated to the collection and analysis of bicycle crash-related videos. It comprises 3,000 dashcam videos, totaling 436,347 frames, and covers a wide spectrum of critical scenarios spanning from collision events to safe interactions. The construction of the CycleCrash dataset involves acquiring videos from multiple online platforms, followed by a suite of preprocessing operations including temporal cropping, file format conversion, spatial cropping and scaling, frame rate adjustment, and color normalization. The CycleCrash dataset is primarily targeted at the autonomous driving and driver-assistance research communities, aiming to address bicycle safety concerns such as collision prediction and time-to-collision prediction by offering detailed annotations for crash incidents, cyclists, and contextual scenes.




