ADNet: Temporal Anomaly Detection in Surveillance Videos

TL;DR

ADNet is a novel neural network architecture utilizing temporal convolutions and a new loss function for effective online anomaly detection in surveillance videos, evaluated on an extended UCF Crime dataset.

Contribution

The paper introduces ADNet with a new loss function and evaluation metric, and extends the UCF Crime dataset with additional classes and annotations.

Findings

ADNet achieves promising results on the extended UCF Crime dataset.

F1@k is proposed as a better evaluation metric than AUC for temporal localization.

The model effectively localizes anomalies in real-time video streams.

Abstract

Anomaly detection in surveillance videos is an important research problem in computer vision. In this paper, we propose ADNet, an anomaly detection network, which utilizes temporal convolutions to localize anomalies in videos. The model works online by accepting consecutive windows consisting of fixed-number of video clips. Features extracted from video clips in a window are fed to ADNet, which allows to localize anomalies in videos effectively. We propose the AD Loss function to improve abnormal segment detection performance of ADNet. Additionally, we propose to use F1@k metric for temporal anomaly detection. F1@k is a better evaluation metric than AUC in terms of not penalizing minor shifts in temporal segments and punishing short false positive temporal segment predictions. Furthermore, we extend UCF Crime dataset by adding two more anomaly classes and providing temporal anomaly annotations for all classes. Finally, we thoroughly evaluate our model on the extended UCF Crime dataset. ADNet produces promising results with respect to F1@k metric. Dataset extensions and code will be publicly available upon publishing