PHEVA
收藏资源简介:
PHEVA数据集由夏洛特北卡罗来纳大学创建,是一个专注于保护隐私的人类中心视频异常检测数据集。该数据集包含超过500万条记录,涵盖了室内外七个不同场景,特别设计了一个专注于执法和安全人员训练的特定情境摄像头。数据集通过去除像素信息,仅提供去标识化的人类注释,如边界框、跟踪ID和人体姿态,以保护个人隐私。PHEVA数据集的创建旨在解决视频监控中异常行为检测的问题,特别是在隐私保护和伦理考虑方面,为持续学习和异常检测算法提供了新的基准。
The PHEVA dataset, created by the University of North Carolina at Charlotte, is a privacy-preserving human-centric video anomaly detection dataset. It contains over 5 million records covering seven distinct indoor and outdoor scenarios, and specially incorporates custom scenario cameras tailored for the training of law enforcement and security personnel. To safeguard personal privacy, the dataset removes pixel-level information and only provides de-identified human annotations including bounding boxes, tracking IDs, and human poses. The creation of the PHEVA dataset aims to address the challenges of abnormal behavior detection in video surveillance, particularly with regard to privacy protection and ethical considerations, and provides a new benchmark for continual learning and anomaly detection algorithms.




