Intelligent Identification And Safety Risk Assessment Of Dangerous Driving Behavior Based On Driver’s Facial Information

When driving a vehicle,in addition to the risk control of the vehicle that requires our attention,the risk control of the driver also deserves our attention.In the vast majority of cases,for the safe driving of drivers,more attention is paid to whether the driver is fatigued,whether there are behaviors such as making phone calls,smoking,etc.,while ignoring the influence of emotions on the driver’s attention and blind driving behavior.hazard caused.The driver’s emotional changes will affect the driver’s attention,which will easily lead to safety accidents.Blind driving behavior means that the eyes’ recognition range is not in front of the road.When encountering an emergency,it is difficult to detect and take emergency measures at the first time.It is easy to cause safety accidents.Because the human face is the most direct platform for human emotion display and sight capture,the expression data and pupil orientation data provided by the human face can be a feasible method to identify and detect the driver’s emotional driving and blind driving behavior.The purpose of this paper is to complete a set of process of facial expression recognition and blind driving behavior detection on the basis of considering various external environments and drivers’ own factors,and to compare the results of facial expression recognition and blind driving behavior detection with the established risk early warning model.Combined,the results returned by the risk warning model can remind and help drivers get rid of dangerous driving behaviors and return to normal.The specific work and results of this paper are as follows:(1)After collecting a series of open source expression data sets and pupil orientation data sets,select some qualified data to make a data set,and take into account the dual variables of the driver’s environment and the driver himself,according to changes in the environment,some data The set performs color change based on PCA principal component analysis and low-exposure data generation based on Cycle GAN cyclic generative network to imitate the changes in the ambient light of the driver.For the driver’s head turning problem,some datasets are pivoted.In this way,the dataset covers as many real situations as possible and enhances the robustness of the model.In the data preprocessing,in order to reduce some irrelevant interference items of background and color,and some data sets are collected in the case of weak light or synthesized through a recurrent neural network,if these data sets are directly transmitted to The neural network will have a certain negative impact on the final result.In order to reduce the influence of non-face facial variables as much as possible,the data is enhanced,including dark light enhancement processing based on the Enlighten GAN network for the collected and generated low-exposure images,and the color and size of the expression recognition algorithm.The normalization processing and face cropping based on dlib face keypoint detection and three court five eyes prior knowledge.(2)After a comprehensive comparison of various classification networks,it is proposed to use the classical residual network to recognize expressions.In order to better deploy on the terminal,model fusion and parameter quantization are performed on the model,and static quantization after training is used to optimize the model.Finally,on the basis of ensuring the accuracy of the model,the model size and operating efficiency are optimized to a certain extent.(3)The functional relationship mapping between the position change of the pupil center point and the line of sight change range under the head upright posture is determined.Taking into account the existing key point detection algorithms,and considering the needs of the actual situation,HRNet is finally selected as the key point detection algorithm,and the results of the key point detection are used as the parameters of the subsequent blind driving warning model.(4)Referring to the basis of risk level classification,comprehensively considering the relevant factors and safety norms involved in drivers’ safe driving,a mathematical model of driver’s emotional driving warning and blind driving warning is established,and verified in an example.