Research On Driver Behavior Detection Method Based On Video

With the improvement of people’s living standards and the development of science and technology,cars have become an important means of transportation for people to travel.With the surge in the number of cars and drivers,some traffic accidents also follow.The accurate and real-time detection of driver’s behavior has a wake-up effect on driver’s driving behavior and is of great significance to driving safety;Therefore,through the video image preprocessing of drivers,this paper mainly solves the problem of low illumination image enhancement in driving environment;The yolov5 target recognition algorithm is used to locate the driver’s face,mouth and eyes;Use pose to locate the key points of the driver model;Finally,deep learning is used to train the samples,and achieve the driver behavior detection.The main work and innovations of this paper are as follows:In the image preprocessing part,RETINEX illumination algorithm is introduced,and the algorithm is improved by parameter adjustment.Compared with the original algorithm,it is found that the improved algorithm has better enhancement effect for low illumination images with face information.It not only takes into account the original image information,but also applies to the driving environment with changeable light.In the aspect of fatigue driving detection,Yolov5 algorithm is applied to the target detection of driving environment.Input ibug-300 w data set for learning,which improves the efficiency of face location and face information feature extraction.Finally,by adjusting the parameters of fatigue discrimination,the judgment of driver fatigue driving is realized,and the driving state of driver is detected in real time.YAW DD fatigue driving data set and self collected driving data set are selected to verify the fatigue driving detection algorithm in this paper.In order to more comprehensively detect the driver’s behavior and warn the dangerous driving state,the OpenPose attitude estimation model is combined with deep learning to realize the driver’s behavior detection software,which can detect six driving actions: driving,drinking,smoking,calling,lower head and look around.Finally,the training sample interface and recognition results are visualized by using python programming language and development tools such as pyqt5.Through the realtime video collected by the camera for learning,the information of bone key points is counted,and the input of a small number of samples can accurately estimate the attitude of the driver.The image enhancement algorithm is applied to driver behavior detection,and the driving behavior detection experiment in low illumination driving environment is carried out.