基于自编码器框架的自适应视频异常检测方法  

Adaptive Anomaly Detection Method in Video Based on Autoenconder Framework

作　　者：蔡德秀 杨大为[1] CAI Dexiu;YANG Dawei(School of Information Science and Engineering,Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学信息科学与工程学院

出　　处：《微处理机》2019年第5期60-64,共5页Microprocessors

摘　　要：为解决视频序列中的异常事件检测问题,提出一种基于自编码器框架的自适应异常检测方法。方法对视频序列进行预处理,将原始数据转化为模型可接受的输入;利用ConvLSTM网络模型构建编码器与解码器,用来学习视频序列的空间特征表示,最小化学习输入视频与输出视频之间的重构误差。由重构误差进行归一化得到规律分数,将其与自适应设置阈值相结合,进行异常事件检测,当规律分数低于自适应设置阈值时检测出视频中的异常。实验结果表明,该方法能够从视频序列中学习规律,自适应地检测视频中的异常事件。In order to solve the problem of anomaly detection in video sequences, an adaptive anomaly detection method based on autoencoder framework is proposed. The method is used to preprocess the video sequence, and the original data was converted into input acceptable to the model. The encoder and decoder are constructed by using the ConvLSTM network model to learn the spatial feature representation of the video sequence and minimize the reconstruction error between the learned input video and the output video. Normalizing the reconstruction error to obtain a regular score, combining the score with an adaptive setting threshold value to detect abnormal events, and detecting abnormalities in the video when the score is lower than the adaptive setting threshold value. Experimental results show that the method can learn rules from video sequences and adaptively detect abnormal events in video.

关 键 词：异常事件检测 ConvLSTM网络 自编码器 自适应 重构误差 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]