Research On Recognition Technology Of Operation Behavior Of Subway Drivers Based On Deep Learning

In the subway operation,whether the subway driver’s operation behavior conforms to the safety standard directly relates to the passenger’s personal safety,so the video monitoring of the subway driver is a very important security measure.As an important application scenario of human behavior recognition,the traditional behavior recognition method needs to design target features by hand and extract features by specific methods,there are problems of low efficiency,low accuracy and poor generalization.Deep convolution neural network has the characteristics of self-learning,which can automatically extract the effective information in the data,and the efficiency and accuracy are greatly improved.In this paper,based on deep learning,the algorithm of operation behavior recognition of Metro drivers is further studied.The main work is as follows:Firstly,the recognition of subway driver’s operation behavior based on two-dimensional convolution neural network is completed.The basic framework uses two-dimensional Inception-v3 as the basic network,constructs a two-stream convolution network model,and extracts different data forms in video images as RGB and optical flow data.optical flow data.The experimental results show that in the two stream network,the recognition accuracy is improved to 81.63% by using part of BN dropout in the training process,and the recognition accuracy is 82.16% by using the average pooling function.Secondly,the recognition of subway driver’s operation behavior based on 3D residual convolution neural network is completed.In order to make full use of the spatial and temporal information,a 3D convolution neural network is designed and used to extract the temporal and spatial features and embed the 3D residual structure.The experimental results show that the recognition accuracy is 83.23% after two fine adjustments(in general scenarios),and the recognition accuracy continues to rise with the increase of the number of layers of convolution at the bottom of the frozen layer,but the convolution at the top can not be arbitrarily frozen.The recognition accuracy of SVM method(in general scenarios)is 82.28%,and the recognition accuracy of using the global average pooling operation instead of the full connection layer is 83.17%.Finally,the recognition of subway driver’s operation behavior based on multi flow convolution neural network is completed.In order to enhance the generalization performance of the network model,the attention mechanism of the two-dimensional convolution neural network and the residual structure of the three-dimensional convolution neural network are combined,using multimodal data input,and finally based on the feature cascade,feature addition two algorithms are compared.The experimental results show that the recognition accuracy of general scene is 86.78%,and that of general scene is 85.47%.The accuracy is high and convenient,so the combination of deep learning and Metro driver operation behavior recognition will have a very broad application prospect.