Retraining Generative Adversarial Autoencoder for Video Anomaly Detection

Video anomaly detection is a challenging task that has received increasing attention in the computer vision community. However, most existing methods train the model only once, leading to misclassification of normal frames during testing and a high equal error rate. To address this problem, a novel retraining generative adversarial autoencoder (RGaAE) method is proposed in this paper. Concretely, during testing, misclassified normal frames are retrained by adding them to the training set, and the model is retrained several times to reduce the probability of normal frames being misclassified. Additionly, a regularization function is added to prevent overfitting during training. Experimental results on three publicly available datasets show that our method effectively reduces the probability of normal frames being misclassified during testing, thereby improving the detection accuracy. Compared to state-of-the-art video anomaly detection methods, our approach has shown obvious superiority in equal error rate.