Self-Trained Video Anomaly Detection Based on Teacher-Student Model

Anomaly detection in videos is a challenging problem in computer vision. Most existing methods need supervised information to train their models, which limits their applications in real world scenario. Therefore, self-Trained methods which do not need manually labels receive increasing attentions recently. In this paper, we propose a novel self-Trained video anomaly detection method based on teacher-student model. The teacher-student architecture can significantly improve the performance of self-Trained video anomaly detection by utilizing the unlabeled samples. We test our method on two surveillance datasets. Experiment results show that our method achieves better performance than state-of-The-Art unsupervised methods on both datasets and achieves comparable performance as semi-supervised methods, which experimentally proves the effectiveness of our method.