VISO
收藏资源简介:
VISO数据集是由吉林一号卫星星座收集的大规模卫星视频数据集,包含47个高质量视频,总计1,646,038个目标实例和3,711个轨迹,用于移动目标检测和跟踪。数据集涵盖多种城市规模元素和移动车辆,如飞机、汽车、船只和火车,适用于城市规模交通管理、海洋监测和智慧城市等应用。创建过程中,研究人员手动标注了所有实例,并使用累积多帧差分和鲁棒矩阵完成技术提高检测率和减少误报。VISO数据集是首个公开的卫星视频移动目标检测和跟踪基准,可用于评估现有检测器和跟踪器的性能,解决卫星视频中移动目标检测和跟踪的挑战。
The VISO dataset is a large-scale satellite video dataset collected by the Jilin-1 satellite constellation. It consists of 47 high-quality videos, containing a total of 1,646,038 target instances and 3,711 trajectories, and is tailored for moving object detection and tracking tasks. The dataset covers various urban-scale elements and moving vehicles including aircraft, cars, vessels and trains, making it suitable for applications such as urban-scale traffic management, ocean monitoring and smart city development. During its construction, researchers manually annotated all instances, and applied cumulative multi-frame difference and robust matrix completion techniques to enhance detection rates and reduce false alarms. The VISO dataset is the first publicly available benchmark for moving object detection and tracking in satellite videos, which can be used to evaluate the performance of existing detectors and trackers, and tackle the challenges in moving object detection and tracking from satellite videos.




