Anomaly detection with false alarm rate controllable classifiers

Pelvan, Soner Özgün and Can, Başarbatu and Özkan, Hüseyin (2023) Anomaly detection with false alarm rate controllable classifiers. In: 31st Signal Processing and Communications Applications Conference (SIU), Istanbul, Turkiye

Full text not available from this repository. (Request a copy)


Automatic anomaly detection gains attention as the number of surveillance cameras increase. Research on the field concentrates on the development of features modeling the overall behavior in a scene, and with each iteration drives the performance to its limit. However, there is a lack of research for detecting anomalous behavior not only based on extracted features, but also using its spatial properties. In this paper, we propose a framework that utilizes Neyman-Pearson classification for detecting anomalies and also providing a true local anomaly detection method, which is capable of detecting same action as normal or anomalous depending on its spatial properties. Our method is a context tree based, competitive ensemble NP classifier, containing multiple piece-wise linear NP models trained on different partitions of a video frame. Even though the number of partitions increases doubly exponentially with the increasing tree size, efficient tree framework provides training in linear time complexity. Competitive nature of the learning of the ensemble model ensures convergence to the optimal space partitioning depending on the performances of individual NP classifiers.
Item Type: Papers in Conference Proceedings
Uncontrolled Keywords: Anomaly detection; False alarm rate; Neyman-Pearson classification; Optimal space partitioning
Divisions: Faculty of Engineering and Natural Sciences > Academic programs > Electronics
Faculty of Engineering and Natural Sciences
Depositing User: Hüseyin Özkan
Date Deposited: 07 Feb 2024 10:59
Last Modified: 07 Feb 2024 10:59

Actions (login required)

View Item
View Item