Study On The Method Of Driving Behavior Recognition Based On Deep Learning

The standardization of driving behavior is not only the guarantee of traffic safety,but also the premise to solve the traffic rush.The research on driving behavior recognition is one of the hotspots in the research of road traffic safety.This paper is based on deep learning algorithm for driving behavior recognition of visual images,and its research has important theoretical significance and application value.The main contents of the paper are as follows:First,based on the basic image processing algorithm,through theoretical analysis and experimental verification,a series of image enhancement,image filtering and descending sampling algorithms were tested,and the basic preprocessing methods based on histogram equalization,median filtering and Gauss Pyramid algorithm were studied.At the same time,the feature pre-extraction algorithm based on Gauss hybrid model was studied,which reduces the impact of background factors on the recognition results and speeds up the convergence speed of the network model.Then,a classification method of driver's upper body posture based on improved LRCN network model was studied,which combines convolution neural network with GRU recursive network,and some image preprocessing algorithms and optimization strategies were added at the same time,and a complete classification model of driver upper body attitude classification is established.This model has high recognition accuracy for daytime and nighttime driver's upper body posture samples.A driver head pose estimation method based on multiple loss fusion network model was proposed.A pre-trained deep convolution network model was used as the backbone network.The classification loss were fused with the regression loss,and the loss value was calculated separately for each angle,then,the model is trained by the method of migration learning.The model can accurately estimate the Euler angle of the face,and provide the basis for segmentation of sub-sequence samples for the upper body pose classification model.Meanwhile,it can correct the results of the upper body attitude classification,and further improve the accuracy of the driving behavior recognition.Finally,aiming at image preprocessing algorithm and several network models,a series of related experiments were designed to test its performance.And the overall design of driving behavior recognition system is studied and expounded.Based on the computer vision technology and the deep learning algorithm,according to the requirement analysis of driving behavior identification,the functional modules of the system were defined and designed,and the overall framework of the driving behavior identification system was determined.The software program of the result record and display module was realized by using the PyQT5 framework in Python language.This program integrates various algorithms and network models in the paper,can identify the input driving behavior images,record and display the recognition results.