I see you
收藏资源简介:
I see you数据集是由圣安东尼奥阿巴德国立大学等机构创建,专注于车辆与行人交互的近事故场景数据。该数据集包含170个近事故事件,来源于秘鲁库斯科市的七个交叉路口的交通监控视频。数据集通过YOLOv5和相机校准方法获取轨迹数据,旨在为自动驾驶车辆提供复杂交互场景下的训练数据,以提高安全性。数据集创建过程中,使用了YOLOv5进行物体检测,StrongSORT进行跟踪,并采用线性卡尔曼滤波器去除轨迹噪声,最终将像素坐标转换为GPS坐标。该数据集适用于研究车辆与行人的交互行为,特别是在近事故场景中的行为预测和风险评估。
The 'I See You' Dataset was developed by institutions including the National University of San Antonio Abad, focusing on near-crash scenario data for vehicle-pedestrian interaction. This dataset comprises 170 near-crash incidents collected from traffic surveillance videos at seven intersections in Cusco, Peru. Trajectory data is acquired via YOLOv5 and camera calibration techniques, with the objective of supplying training datasets for autonomous vehicles operating in complex interaction scenarios to improve traffic safety. During the dataset construction pipeline, YOLOv5 is utilized for object detection, StrongSORT for multi-object tracking, and a linear Kalman filter is applied to remove trajectory noise; pixel coordinates are finally converted to GPS coordinates. This dataset is suitable for research on vehicle-pedestrian interaction behaviors, particularly behavior prediction and risk assessment in near-crash scenarios.




