USVD (Urban Surveillance Video Dataset)
收藏资源简介:
USVD是由中国科学技术大学电磁空间信息重点实验室和中国科学院创建的大型城市监控视频数据集,旨在支持多目标跟踪和行为分析研究。该数据集包含16个场景,覆盖7种典型的户外环境,如街道、交叉路口、医院入口等。数据集包含超过20万帧视频,标注了超过370万个对象边界框和约7100条轨迹。创建过程中,所有边界框均经过人工标注和检查,确保了数据的质量。USVD数据集适用于评估城市拥挤场景下的多目标跟踪和异常行为检测算法,为公共安全和城市管理提供技术支持。
USVD is a large-scale urban surveillance video dataset developed by the Key Laboratory of Electromagnetic Space Information, University of Science and Technology of China and the Chinese Academy of Sciences, aiming to support research on multi-object tracking and behavior analysis. This dataset includes 16 scenarios covering 7 typical outdoor environments such as streets, intersections, hospital entrances and other similar places. It contains over 200,000 video frames, with more than 3.7 million annotated object bounding boxes and approximately 7,100 trajectories. During the dataset construction process, all bounding boxes were manually annotated and verified to ensure data quality. The USVD dataset is suitable for evaluating multi-object tracking and abnormal behavior detection algorithms in crowded urban scenes, providing technical support for public safety and urban management.




