显著性光流直方图字典表示的群体异常事件检测

发布时间：2018-11-15 08:29

【摘要】：在视频监控系统已被广泛应用的今天,基于监控视频的群体异常事件检测已成为保障社会安全的迫切需要,越来越受到人们的重视。该文基于这一现状,提出了一个新的群体异常事件检测方案,实现对监控视频自动高效的检测。在特征提取方面,提出了显著性光流直方图特征描述符,并利用该特征描述符构建字典;在字典优化方面,提出了基于聚类的多字典组合学习框架,将原始的大字典分为多个子字典;最后,对于测试样本,找出最适合的子字典并计算测试样本在该子字典下的重建误差,即可判断测试样本是否异常。在两个数据集上的实验表明,与其他方法相比,该文提出的方法对拥挤场景下监控视频中的群体异常事件检测取得了较好的检测性能。
[Abstract]:Nowadays, video surveillance system has been widely used. Group abnormal event detection based on video surveillance has become an urgent need to ensure social security, and has been paid more and more attention. Based on this situation, a new group anomaly detection scheme is proposed in this paper, which can detect surveillance video automatically and efficiently. In the aspect of feature extraction, a significant optical flow histogram feature descriptor is proposed and used to construct a dictionary. In the aspect of dictionary optimization, a multi-dictionary combinatorial learning framework based on clustering is proposed, which divides the original large dictionary into several sub-dictionaries. Finally, for the test sample, find out the most suitable sub-dictionary and calculate the reconstruction error of the test sample under the sub-dictionary, we can judge whether the test sample is abnormal or not. Experiments on two datasets show that compared with other methods, the method proposed in this paper achieves better performance in detecting crowd abnormal events in overcrowded video.
【作者单位】：北京交通大学信息科学研究所;现代信息科学与网络技术北京市重点实验室;广东财经大学电子商务研究院;北京建筑大学理学院;
【基金】：国家自然科学基金(61572067,61502024) 北京市自然科学基金(4162050) 广东省自然科学基金(2016A030313708) 中央高校基本科研业务费(K16JB00110)
【分类号】：TP391.41;TN948.6

【相似文献】