Research On Driver Risky Behavior Recognition Method Based On Deep Learning

With the rapid growth of car resource holdings and the frequent occurrence of traffic accidents,research on driving safety and other related issues has also continued to deepen.Image processing technology based on deep learning can not only complete target detection,target recognition and image feature extraction,but also provide new intelligent solutions and means for driver's dangerous behavior recognition and fatigue detection.As an important direction of artificial intelligence,deep learning technology can obtain the inherent laws and essential characteristics of data samples through a series of loop iterations,and has great application and development potential for artificial intelligence.This subject mainly studies the method of identifying dangerous behaviors of drivers based on deep learning.Combining machine learning and basic theories related to deep learning,the research focuses on the recognition algorithms of dangerous behavior recognition of drivers based on machine learning and deep learning,and the Driver fatigue detection and recognition algorithm.First,from the three image filtering algorithms of smoothing filtering,Gaussian filtering,and median filtering,the driver behavior data set and the driver fatigue detection data set are preprocessed,and different types of preprocessing algorithms are compared and analyzed;further study of image features The acquisition algorithm deeply explores the calculation process of the extraction algorithm of LBP and HOG features of driving images,and further studies the three types of driver behavior based on LBP-SVM,HOG-SVM and feature fusion-SVM with the help of support vector machines Recognize the algorithm,and complete the simulation experiment of the three driver recognition algorithms.Second,explore the basic ideas of deep learning,discuss the internal logic of deep learning to deal with image recognition problems;study the basic composition of neural networks,and discuss the comparison of different activation functions,forward propagation algorithms,back propagation algorithms,and various optimizations The evolution and optimization process of the algorithm;focus on the principle of convolutional neural network,describe the internal calculation process,and explore the back propagation process of the convolutional layer and the pooling layer for the two unique structures of the convolutional layer and the pooling layer.Subsequent implementation of algorithms based on deep learning provides a certain degree of theoretical support;Then,on the basis of the above neural network theory research,explore the driver behavior recognition algorithm based on neural network,and simulate and complete three driver behavior recognition experiments based on LBP-NN,HOG-NN and feature fusion-NN;in the above Based on the theoretical research of the convolutional neural network,explore the driver behavior recognition algorithm based on the convolutional neural network and complete the corresponding simulation experiment;on the basis of the convolutional neural network experiment,explore the CNN improved algorithm and complete the corresponding driver Behavior recognition simulation experiment;on the basis of improved algorithms of HOG-SVM,CNN,CNN,explored the driver behavior recognition algorithm based on multi-decision fusion,and obtained the best driver behavior recognition experiment results in this paper.Finally,on the basis of HOG theoretical research,explore the recognition algorithm based on HOG features,and carry out the face feature point extraction experiment based on the cascade regression tree for the detected face slices to extract 68 feature points of the face;and According to the two indicators of eye aspect ratio and mouth opening,respectively carry out fatigue detection experiment based on eye aspect ratio and fatigue detection experiment based on mouth opening;explore the principle of multidimensional information perception,explore fatigue detection based on multidimensional information perception experiment.