Research On Driving Safety Assessment Method Based On Deep Learning

With the improvement of people’s living standards,cars bring convenience to people’s lives,but also result in a lot of traffic accidents.According to the analysis of the accident causes,lots of traffic accidents caused by fatigue driving are no less than drunk driving.How to obtain the drivers’fatigue characteristics so as to evaluate the degree of their fatigue has become a hot issue at home and abroad.In current research on driving fatigue detection,most of them are characterized by extracting and analyzing eyes and pupils features.However,usually the fatigue state also causes the changes in the set of mouth,such as yawning.Therefore,it is also of great significance to detect the state of the mouth.This paper proposes to study the single feature recognition fatigue between eyes and mouth and the fusion of the two depth features to identify the fatigue.It is proposed to use deep learning instead of the traditional machine learning method to complete the fatigue feature extraction,which has higher accuracy and robustness.The main research contents of the thesis are as follows:(1)Face detection and facial key point positioning are the bases for studying and analyzing driving safety assessments.Considering the real-time and accuracy of the algorithm,the Adaboost algorithm is used for face detection.The millisecond-level face alignment algorithm based on the regression tree is used to locate the key points of detected facial data.Eyes and mouth regions are extracted and segmented respectively according to the location of eight feature points detected.(2)Aiming at the mouth extracting feature is still a difficult problem at present.It is proposed to use the convolutional neural network to identify the state of the mouth improving the robustness and recognition rate of the algorithm.And,it is also proposed a method to identify the fatigue state of the mouth.The experimental results show that the state recognition and fatigue detection algorithms both have higher accuracy.The convolutional neural network is also used to identify the state of eyes.And the fatigue state is identified based on the current PERCLOS parameter that makes the best effective judgement on driver’s fatigue.(3)Aiming at the problem of eyes and mouth feature fusion recognition fatigue,a fatigue recognition model of convolutional neural network and the coding vector is put forward.The convolutional neural network model is used to extract and classify each frame image.And the code vectors are classified by Softmax regression so that the fatigue state of the driver in the video can be detected.The experimental results show that the fatigue detection based on deep learning framework can effectively improve the robustness and accuracy of the algorithm.