JAAD

JAAD是在自动驾驶的背景下研究共同注意力的数据集。重点是过马路时的行人和驾驶员行为以及影响他们的因素。为此，JAAD数据集提供了从超过240小时的驾驶镜头中提取的346短视频剪辑 (5-10秒长) 的丰富注释集合。这些在北美和东欧的几个地点拍摄的视频代表了在各种天气条件下日常城市驾驶的典型场景。为所有行人提供了带有遮挡标签的边界框，使该数据集适合行人检测。行为注释为与驾驶员互动或需要驾驶员注意的行人指定了行为。对于每个视频，都有几个标签 (天气，位置等) 和固定列表中的带时间戳的行为标签 (例如，停止，行走，观看等)。此外，为每个行人提供人口统计属性列表 (例如年龄、性别、运动方向等) 以及每个帧的可见交通场景元素列表 (例如停车标志、交通信号等)。JAAD is a dataset for researching joint attention in the context of autonomous driving, focusing on the behaviors of pedestrians and drivers during road crossing and the factors affecting them. To this end, the JAAD dataset provides a rich annotated collection of 346 short video clips (5–10 seconds in length) extracted from over 240 hours of driving footage. These videos, shot at multiple locations in North America and Eastern Europe, represent typical scenarios of daily urban driving under various weather conditions. Bounding boxes with occlusion labels are provided for all pedestrians, making this dataset suitable for pedestrian detection tasks. Behavior annotations assign behaviors to pedestrians who interact with drivers or require drivers' attention. For each video, there are several tags (e.g., "weather", "location") and timestamped behavior labels from a fixed list (e.g., "stopping", "walking", "looking", etc.). In addition, a list of demographic attributes (e.g., "age", "gender", "movement direction", etc.) for each pedestrian and a list of visible traffic scene elements (e.g., "stop signs", "traffic signals", etc.) per frame are provided.

搜集汇总

背景与挑战

背景概述

JAAD是一个专注于自动驾驶场景下共同注意力研究的数据集，包含从240多小时驾驶视频中提取的346个短视频，覆盖北美和东欧的多种天气条件。数据集提供了丰富的注释，包括行人检测边界框、行为标签、人口统计属性和交通场景元素，适用于行人检测和行为分析任务。