Method for detecting riot and terror videos based on CNN and LSTM

Abstract

The invention discloses a method for detecting riot and terror videos based on CNN and LSTM, and belongs to the technical field of pattern recognition, video detection and deep learning. The detection method comprises following steps: firstly, key frame sampling is performed on the video to be detected and key frame features are extracted; expression and discrimination at video level are performed, wherein the expression and the discrimination comprise VLAD feature expression and SVM discrimination of a CNN semantic module, VLAD feature expression and SVM discrimination of a CNN scene module, and LSTM discrimination of a LSTM time sequence module; finally, results are fused. According to the method, the advantages of CNN on image feature extraction and LSTM on time sequence expression are utilized, and features of riot and terror videos at scene are taken into full consideration; test index mAP value reaches 98.0% in real tests which approaches manual operation level. In terms of operation speed, only single machine GPU acceleration mode is adopted and 76.4 seconds of network video can be processed per second; the method is suitable for blocking the spread of riot and terror videos on large video websites and therefore it helps maintain social stability and state long-term peace and order.

Description

Technical field [0001] The invention belongs to the technical fields of pattern recognition, video detection, and deep learning, and specifically relates to a CNN and LSTM-based violent terror video detection method. Background technique [0002] In recent years, a large number of violent terrorist videos at home and abroad have been illegally spread on the Internet, and they have become a major cancer that endangers social stability. However, related automated violent terrorist video detection technologies are still in the research and development stage. Most of them use existing event video detection methods. These methods can basically be divided into three categories: video detection methods based on image local features, video detection based on semantic concepts Method and video detection method based on Convolutional Neural Network (CNN). [0003] Reference[1](Sun,Chen,and Ram Nevatia."Large-scale web video event classification by use of fisher vectors."In Applications of C...

AI Technical Summary

Problems solved by technology

This method does not require too much manual labeling in the training process, and is simple and easy to implement, but it has the following shortcomings: (1) The detection accuracy is limited by the local features used

(2) The detection speed is slow

Due to the orientation of the preset semantic concepts, the video detection method based on semantic concepts has higher accuracy for violent and terrorist video recognition, but has the following disadvantages: (1) A large number of labeled image samples are required in the training process, and the labor cost is relatively large

(2) When the violent and terrorist video to be detected does not contain any preset concept, the detection accuracy is not guaranteed

(3) The detection speed is slow

This method makes full use of the advantages of CNN in static image feature extraction, and can achieve better results in violent video detection, but there are still the following aspects that can be improved: (1) The method is not difficult to detect in the process of VLAD feature expression. The timing characteristics of the video are not fully utilized

(2) This method only extracts the CNN semantic features of key frames, and does not pay attention to other personalized features of violent terror videos

In summary, there is still room for performance improvement in video detection methods based on CNN semantic features

Images