A method and system for abnormal event detection based on attention mechanism

Abstract

A method and system for monitoring abnormal events based on an attention mechanism, including a model training step and a data processing step, and the model training step includes a forward propagation step and a backward propagation step. In the forward propagation step, the VGG network structure is selected, and the selected pictures are extracted through the VGG network. Each picture gets a k*k*D feature cube. In the attention-based model, each time, LSTM will generate a k The probability value of the *k area is multiplied with the feature cube of the next frame as the input of the next frame. When the backward propagation step is trained with training data, when the training of the previous frame is completed, the optimal solution of the double random penalty function formed by the loss function and the attention penalty is used to update the weight value of the forward propagation. Data processing step: use the trained model to process the video to be processed, and detect abnormal events and abnormal event occurrence areas. The present invention can not only better detect the abnormality but also detect the range of the abnormal area.

Description

technical field [0001] The present invention relates to the field of pattern recognition, and in particular to abnormal event detection algorithms based on depth expression. More specifically, the present invention relates to a method and system for detecting abnormal events based on attention mechanism. The attention area probability obtained by the network (Long short-term memory, referred to as LSTM) and the features extracted by VGG detect the position of the abnormal area. Background technique [0002] With the advancement of projects such as "Safe City" and "Safe Campus", intelligent video surveillance has become an important technology in the field of public security, and abnormal event detection is a major way to improve the intelligence of video surveillance. Abnormal event detection aims to analyze the content of events from a large amount of video data, summarize the rules of normal events, and detect abnormal video events that are different from them. Effective ...

AI Technical Summary

Problems solved by technology

(3) Unpredictability

A large number of studies have shown that the features of video events are very important to the accuracy of anomaly detection, and the existing hand-designed features are not suitable for all situations. Currently, the deep learning features that are widely used in the field of computer vision and are very effective in abnormal video The field of event detection is rarely covered

Due to the temporal characteristics of video events, the video content between adjacent frames is related, and the hidden layers of ordinary deep neural networks are independent, and cannot mine the temporal variation characteristics of events.

Moreover, it can only detect abnormalities at present, but cannot detect the location of abnormal areas.

Images