Real-World Video Anomaly Detection by Extracting Salient Features

We propose a lightweight and highly accurate method for detecting anomalies in videos. Existing methods use multiple-instance learning (MIL) to determine the normal/abnormal status of each segment of the video. Recent successful researches argue that it is important to learn the temporal relationships among segments to achieve high accuracy, instead of focusing on only a single segment. We analyzed the existing methods that have been successful in recent years, and found that while it is indeed important to learn all segments together, the temporal relationships among them are irrelevant to achieving high accuracy. Based on this finding, we do not use the MIL framework, but instead introduce a self-attention mechanism to automatically extract features that are important for determining normal/abnormal from all input segments. As a result, the neural network with 1.3% of the number of parameters of the existing method can achieve the comparable or better accuracy than the existing method.