Monitoring image anomaly detection based on lightweight network MobileNet

Abstract

At present, there are problems in remote video surveillance that it is difficult to detect image anomalies in complex scenes, and traditional algorithm functions are single (only for a certain specific scene or a certain type of abnormal image). Moreover, the currently commonly used convolutional neural network is deployed on the monitoring device, there are problems such as insufficient device memory and low computing power. In response to the above problems, this paper proposes a recognition method that applies the MobileNet convolutional neural network to the detection of abnormal images in surveillance video. The off-line enhancement method is used to reverse and translate the abnormal images in the surveillance video. The data set can be reasonably expanded, the problem of data imbalance can be solved, and the detection accuracy and generalization performance of abnormal images can be improved. The results show that this method can detect abnormalities such as freeze, overexposure, blur, and colorific distortion in remote video surveillance, with an accuracy rate of 91.75%, and can reduce the amount of model parameters and facilitate model deployment.