Data collection

To construct the dataset, we collect videos from two sources. On one hand, we hire some volunteers to perform various actions at the edges of buildings (e.g. near windows and rooftop guardrails) and shoot videos of them. These videos are captured in tilt upperward angles using wide dynamic range cameras, equipped with CMOS sensors and dot matrix LED infrared lamps. On the other hand, to increase the diversity of scenes, we also collect videos from public websites, e.g. YouTube and Baidu, where the videos have the characteristics that a person appears at the edges of buildings shot in tilt upperward angles. The data is comprised of news video clips and user-generated videos.

Data Annotation

For each person's area in an image, two bounding boxes are annotated, one for the visible region of the person's body and the other for the full body.

Maintenance

Our data will be maintained for a long time, and we will check the accessibility of our dataset regularly. The first author will provide support and can be contacted via e-mail at gaozitao@whu.edu.cn.