Smoking action recognition method based on double-flow convolutional neural network and SVM

Abstract

The invention discloses a dangerous behavior automatic identification method based on a double-flow convolutional neural network. Time features and space features are learned respectively through twodifferent convolutional neural networks, and an action identification result is acquired by using Softmax. Softmax classification results of the two networks are not directly and simply weighted and fused, the outputs of the two training Softmax are used as new input features to be used as an SVM classifier, and finally automatic recognition of the smoking action of the person in the video is achieved. SVM is used for modeling the classification results of the time domain convolution network and the space domain convolution network, and the accuracy of automatic recognition of the smoking behavior of the person based on the video is further improved.

Description

technical field [0001] The present invention relates to human body behavior recognition based on a double-stream convolutional neural network, and more specifically relates to an automatic smoking action recognition method for monitoring video data. Background technique [0002] Smoking is not only a bad habit that is harmful to human health, but also a major cause of safety hazards. For gas stations, oil depots, chemical drug depots and other places where a large amount of flammable and explosive materials are stored, the open flame of smoking may cause safety accidents such as fire or explosion, causing huge economic losses and casualties. At present, the prevention of smoking behavior in the above-mentioned places mainly relies on warning slogans or the supervision of security personnel. It is unreliable to rely on the quality of personnel who may cause huge disasters to be awakened by propaganda slogans, and it is also difficult for security personnel to observe all in t...

AI Technical Summary

Problems solved by technology

Aiming at the problem that it is difficult to extract ideal smoke features for smoking behavior recognition in an open-air environment, the smoking behavior is recognized by recognizing the actions of people, and two different convolutional neural networks are used to learn time features and spatial features respectively, and Softmax is used to obtain action recognition Results: In view of the problem that the similarity of smoking actions is more prone to misjudgment, the Softmax classification results of the two networks are not directly weighted and fused, and the output of the two training Softmax is used as a new input feature to the SVM classifier to improve video quality. Accuracy of recognition of smoking action of Chinese characters

Images