To improve the accuracy of video anomaly detection,an autoencoder algorithm based on mixed attention was proposed.To solve the problem that the powerful generalization ability of autoencoder network may reconstruct abnormal behavior,a hy-brid attention module(CSCFAM)was proposed and fused into the jump connection layer between encoder and decoder to limit the generation of abnormal behavior.To consider the diversity of normal samples,Memory module was introduced at the bottle-neck between encoder and decoder to record the prototype pattern of potential characteristics of normal samples.Experimental results show that the frame-level AUC of the proposed algorithm on UCSD Ped2 and CUHK Avenue data sets reaches 97.3％and 87.0％,respectively.Compared with the current advanced video anomaly detection algorithms,the anomaly detection ability is effectively improved.
关键词
视频异常检测/自编码器/跳跃连接/混合注意力模块/存储记忆模块/异常行为/原型模式
Key words
video anomaly detection/autoencoder/skip connection/hybrid attention module/memory module/abnormal beha-vior/prototype pattern
维普
万方数据