Dangerous behavior automatic identification method based on double-flow convolutional neural network

Abstract

The invention discloses a dangerous behavior automatic identification method based on a double-flow convolutional neural network. According to the method, the influence of a video background on personbehavior identification is reduced by carrying out partial manual annotation on persons in a video, and time features and spatial features in the video are learned by using a LeNet-5 network, and thefused space-time features are sent into a 3D convolutional neural network to complete identification of character actions in the video. Aiming at a large amount of irrelevant background information existing in a video, the method carries out manual marking on figures in a part of video frames, reduces noise interference by adding input supervision information, and effectively solves the problem that the video irrelevant background information interferes with figure action recognition. According to the automatic dangerous behavior recognition method based on the double-flow convolutional neural network and the 3D convolutional neural network, an automatic human dangerous action recognition network is constructed, a human dangerous action video data training network is used, and an automatic human dangerous action recognition model is constructed.

Description

technical field [0001] The present invention relates to a human body pose estimation and behavior recognition based on a dual-stream convolutional neural network, and more specifically relates to a method for automatic recognition of dangerous behaviors for monitoring video data. Background technique [0002] For some specific places such as prisons and banks, it is necessary to avoid violent conflicts, illegal break-ins and other unstable events, so it is necessary to maintain round-the-clock supervision. It takes a lot of time and labor costs to simply inspect the above-mentioned specific places through security personnel, and the efficiency is relatively low. With the popularization of surveillance video, the above-mentioned specific places began to use the combination of surveillance cameras and security personnel, that is, security personnel use surveillance cameras to realize online inspections, and then manually inspect and deal with abnormal areas. This method requi...

AI Technical Summary

Problems solved by technology

Convolutional Neural Networks (CNNs) is currently the mainstream deep learning network in the field of image recognition, but its application in the field of video-based character action recognition has been inhibited. The main reason is that compared with image data sets, video The data set is generally small in size and low in quality (contains a large amount of irrelevant noise); on the other hand, the traditional convolutional neural network cannot fully learn temporal features and cannot effectively extract the association between consecutive frames

Images