A video anomaly detection method, medium and device based on a full attention memory module

Abstract

The application discloses a video anomaly detection method based on a full attention memory module, a medium and equipment, the method preprocesses input data through a hybrid encoder and updates a memory module using a full attention mechanism, adopts the hybrid encoder to preprocess first video feature information extracted by an I3D model, can make the extracted video features more comprehensive, avoid losing valuable information, and adopts the attention mechanism to replace the Top-k selection mechanism, can realize iterative updating of various parameters in the more comprehensive memory module, and effectively solves defects in the existing weak supervision video anomaly detection method based on the memory module.

Images