UCSD (UCSD Anomaly Detection Dataset)

“UCSD 异常检测数据集是使用安装在高处的固定摄像机获取的，可以俯瞰人行道。人行道上的人群密度是可变的，从稀疏到非常拥挤。在正常设置中，视频仅包含行人. 异常事件是由于以下原因之一： 人行道中非行人实体的流通 异常行人运动模式 常见的异常包括骑自行车的人、溜冰者、小推车以及穿过人行道或在人行道周围的草地上行走的人。还记录了坐在轮椅上的人。所有异常都是自然发生的，即它们不是为了组装数据集而上演的。数据被分成 2 个子集，每个子集对应一个不同的场景。从每个场景记录的视频片段被分割成大约 200 帧的各种剪辑。”

The UCSD Anomaly Detection Dataset was captured using fixed cameras mounted at elevated positions overlooking sidewalks. The pedestrian density on the sidewalks varies widely, ranging from sparse to extremely crowded. In normal scenarios, the videos only contain pedestrians. Abnormal events occur due to one of the following reasons: circulation of non-pedestrian entities on sidewalks, and abnormal pedestrian movement patterns. Common anomalies include cyclists, skaters, wheeled carts, as well as people walking across sidewalks or on grass areas adjacent to sidewalks. Individuals using wheelchairs are also recorded. All abnormal events occur naturally, meaning they were not staged specifically for the purpose of compiling this dataset. The dataset is split into two subsets, each corresponding to a distinct scenario. Video footage recorded from each scenario is divided into various clips of approximately 200 frames each.

搜集汇总

背景与挑战

背景概述

UCSD异常检测数据集是一个用于视频异常检测的公开数据集，通过固定摄像机拍摄人行道场景，包含正常行人活动和自然发生的异常事件，如非行人实体或异常运动模式。数据集分为两个子集，每个子集对应不同场景，视频剪辑约200帧，适用于计算机视觉中的异常检测研究。