Street Scene

街景数据集由 46 个训练视频序列和 35 个测试视频序列组成，这些视频序列取自静态 USB 摄像头，俯视有自行车道和人行道的两车道街道场景。有关数据集中的典型帧，请参见图 1。在连续两个夏天的不同时间从相机中收集了视频。所有视频都是白天拍摄的。该数据集具有挑战性，因为发生的活动多种多样，例如汽车驾驶、转弯、停车和停车；步行、慢跑和推婴儿车的行人；和骑在自行车道上的骑自行车的人。此外，视频中还包含不断变化的阴影和移动的背景，例如旗帜和随风飘扬的树木。

The street view dataset comprises 46 training video sequences and 35 test video sequences, which are captured by a stationary USB camera overlooking a two-lane street scene with bike lanes and sidewalks. Refer to Figure 1 for a typical frame from this dataset. The videos were collected at different time points across two consecutive summers, and all were shot during daylight hours. This dataset is challenging due to the diverse activities present, including cars driving, turning, parking and stopping; pedestrians walking, jogging and pushing strollers; and cyclists riding on the bike lanes. Additionally, the videos contain varying shadows and dynamic background elements such as flags and trees swaying in the wind.

搜集汇总

背景与挑战

背景概述

Street Scene是一个用于视频异常检测的计算机视觉数据集，包含46个训练视频序列和35个测试视频序列，视频采集自静态USB摄像头，拍摄于两车道街道场景，涵盖汽车、行人、骑自行车者等多种活动以及阴影和背景变化等挑战性因素。该数据集由三菱电机研究实验室和北卡罗来纳州立大学于2020年发布，旨在为视频异常检测任务提供评估基准。