MEVA(Multiview Extended Video with Activities)

用于人类活动识别的大规模数据集。现有的安全数据集要么通过聚合因其内容而传播的公共视频（通常不包括同场景背景视频）来关注活动计数，要么通过观察公共区域来实现持久性，因此无法控制活动内容。该数据集包含超过 9300 小时的未修剪的连续视频，脚本包括各种同步活动以及自发的背景活动。

A large-scale dataset for human activity recognition. Existing security-focused datasets either prioritize activity counting by aggregating public videos circulated for their content (usually excluding background videos from the same scene) or enable persistent monitoring by observing public areas, thus failing to control the content of the recorded activities. This dataset contains over 9,300 hours of untrimmed continuous footage, with its scenarios encompassing various synchronous activities as well as spontaneous background activities.

搜集汇总

背景与挑战

背景概述

MEVA是一个用于人类活动识别的大规模数据集，包含超过9300小时的未修剪连续视频，涵盖了同步脚本活动和自发的背景活动。该数据集通过控制活动内容，弥补了现有安全数据集在活动计数和公共区域持久性方面的局限性。