lv

近年来，设计和测试视频异常检测方法的重点是合成或不真实的序列。这主要有四个缺点：1）事件是可控的和可预测的，因为它们通常由演员执行； 2) 环境条件，例如摄像机运动和照明通常是理想的，因此不能很好地反映现实条件； 3) 事件通常很短且重复； 4) 从不一定与测试场景匹配的场景中捕获材料。这导致我们提出了一个新的、丰富的、由监控摄像机在具有挑战性的环境条件下捕获的逼真视频集合，即实时视频 (LV) 数据集。我们探索了一些最先进的视频异常检测方法在 LV 数据集上的性能。我们的结果证实，需要设计能够以可接受的处理时间处理监控摄像机捕获的逼真视频的方法。因此，提议的 LV 数据集将有助于此类新方法的设计和测试。In recent years, research on designing and testing video anomaly detection methods has focused on synthetic or unrealistic sequences. This has four major drawbacks: 1) Events are controllable and predictable, as they are typically performed by actors; 2) Environmental conditions such as camera motion and lighting are often ideal, and thus do not accurately reflect real-world scenarios; 3) Events are usually short and repetitive; 4) Materials are captured from scenarios that do not necessarily match the test environments. This motivated us to propose a new, rich collection of realistic videos captured by surveillance cameras under challenging environmental conditions, namely the Live Video (LV) dataset. We evaluated the performance of several state-of-the-art video anomaly detection methods on the LV dataset. Our results confirm that there is a need to design methods that can process realistic videos captured by surveillance cameras within acceptable processing times. Thus, the proposed LV dataset will facilitate the design and testing of such novel methods.

搜集汇总

背景与挑战

背景概述

lv数据集是一个用于视频异常检测的公开数据集，包含由监控摄像机在复杂环境条件下捕获的逼真视频，旨在解决现有方法在合成序列上的局限性。该数据集由华威大学和查尔斯特大学于2017年发布，以促进新方法的设计和测试。