PseudoBound: Limiting the anomaly reconstruction capability of one-class classifiers using pseudo anomalies

Due to the rarity of anomalous events, video anomaly detection is typically approached as one-class classification (OCC) problem. Typically in OCC, an autoencoder (AE) trained reconstruct normal only training data with expectation that, test time, it can poorly data. However, previous studies have shown even data, AEs often well, resulting a decreased performance. To mitigate this problem, we propose limit reconstruction capability by incorporating pseudo anomalies during AE. Extensive experiments using five types show robustness our mechanism towards any kind anomaly. Moreover, demonstrate effectiveness proposed based approach against several existing state-of-the-art (SOTA) methods on three benchmark datasets, outperforming all other reconstruction-based approaches two datasets and showing second best performance dataset.