Detection of Traffic Anomaly (DoTA)
收藏资源简介:
DoTA数据集是由密歇根大学和印第安纳大学合作创建的大规模交通视频异常检测数据集,包含4677个视频,用于支持视频异常检测(VAD)和视频动作识别(VAR)研究。该数据集提供了详细的时间、空间和类别标注,旨在通过一个'何时-何地-什么'的分析框架来检测、定位和识别驾驶视频中的异常事件。DoTA数据集不仅规模庞大,而且是首个支持全面交通异常研究的数据集,涵盖了时间、空间和类别的多维度分析。数据集的创建过程涉及从YouTube收集视频,并通过自定义工具进行标注,确保了标注的高质量和一致性。DoTA数据集的应用领域广泛,主要用于解决交通视频中的异常检测问题,支持高级驾驶辅助系统(ADAS)的研发。
The DoTA dataset is a large-scale traffic video anomaly detection dataset co-created by the University of Michigan and Indiana University, consisting of 4,677 videos, designed to support research on video anomaly detection (VAD) and video action recognition (VAR). This dataset provides detailed temporal, spatial, and categorical annotations, aiming to detect, localize, and identify abnormal events in driving videos via a 'when-where-what' analytical framework. Not only is the DoTA dataset large in scale, but it is also the first dataset that supports comprehensive traffic anomaly research, covering multi-dimensional analyses across temporal, spatial, and categorical dimensions. The creation of the dataset involves collecting videos from YouTube and annotating them with custom-built tools, ensuring high quality and consistency of the annotations. The DoTA dataset has a wide range of application scenarios, mainly used to address anomaly detection issues in traffic videos and support the development of advanced driver assistance systems (ADAS).

- 1When, Where, and What? A New Dataset for Anomaly Detection in Driving Videos密歇根大学 · 2020年



