Main-Auxiliary Aggregation Strategy for Video Anomaly Detection

The aim of surveillance video anomaly detection is to detect events that rarely or never happened in a certain scene. Generally, different detectors can anomalies. This paper proposes an efficient strategy aggregate multiple detectors. First, the aggregation chooses one detector as master by experience, and sets remaining auxiliary Then, extracts credible information from detectors, including abnormal (Cred-a) frames normal (Cred-n) frames. After that, frequencies each frame being judged Cred-a Cred-n are counted. Applying events' time continuity property, more be inferred. Finally, utilizes vote calculate soft weights, uses weights assist detector. Experiments carried out on datasets. Comparing with existing strategies, proposed achieves state-of-the-art performance.