Application Research of Monitoring Abnormal Behavior Analysis and Detection Method Based on GANomaly Network
Ning Cao, Yimin Luo · 2022 IEEE 2nd International Conference on Data Science and Computer Application (ICDSCA) · 2022
With the wide application of video surveillance, video surveillance has achieved full coverage of various key places. For video surveillance, the amount of internal information is huge, and the degree of reduction of events is high. Traditionally, a real-time manual inspection can neither meet the full coverage of the time nor achieve the full range of space with low labor cost. The idea of using semi-supervised generative adversarial networks for abnormal behavior detection is proposed to deal with this problem. This paper focuses on this problem, starting from the structure and optimization of the auto-encoder, and studies the basic method of using the encoder for feature extraction. After that, the convolutional generative adversarial network was constructed, and GANomaly was further optimized to detect anomalies by comparing the feature differences between the original image and the abstract image to improve its anti-interference ability. Finally, this paper discusses the reconstruction method of the loss function and defines the loss functions of the adversarial training and discriminative models, respectively. The research in this paper is expected to effectively improve the effect of monitoring abnormal behavior detection and enhance the application value of video surveillance real-time detection.