Research And Implementation Of Unsafe Driving Behavior Based On Machine Vision

In the document No.115^[1]issued by the General Office of the Ministry of Transport,it is pointed out that the driver’s behavior of taking and holding a hand-held phone during driving,long time not looking ahead and fatigue driving will greatly affect the safety of driving.At present,there are relatively few studies on the driving behavior detection of drivers.Existing methods generally use mobile phone signals to monitor the driver’s calling behavior,and deep learning algorithms to estimate the head posture,and then monitor whether the driver looks ahead for a long time and monitor fatigue driving by recording the driving time.Because it is difficult to distinguish whether the driver is making a call based on mobile phone signal detection,the deep learning-based method has high requirements on hardware detection equipment,the model training time is long,and the fatigue detection method based on driving time has the disadvantages of inaccurate detection.This paper proposes the above three aspects of behavior monitoring methods and implementations based on machine vision,and performs classification detection for different environmental scenarios.While realizing all-weather monitoring,improve the accuracy of detection.Firstly,this paper uses the HSV color space algorithm to distinguish the driving environment scene.And for day and night driving scenarios,ordinary color cameras and850nm near-infrared cameras were used to complete the image acquisition and analysis of the driver.Secondly,under all-weather driving conditions,this paper uses driver face detection based on Convolutional Neural Network（MTCNN）framework.For the detection of the driver’s hand-held phone call,through the positioning of the driver’s ear area,the YCb Cr skin color segmentation method and the continuous time monitoring method are used to complete the relevant detection under the daytime driving conditions.Due to the unsatisfactory skin color segmentation effect of near infrared images,in order to complete the relevant detection under night driving conditions,this paper adopts the method of searching for the maximum contour of the ear in continuous time.It is aimed at monitoring the driver for a long time without looking ahead.Based on the detected face feature points,through three-dimensional affine transformation.Obtain the driver’s head posture estimation,and use the abnormal head posture estimation in continuous time to complete the monitoring of the driver’s non-visual front.For the detection of driver fatigue driving,the eyes and mouth of the driver’s face are extracted as the characteristics of driving fatigue.A related eye and mouth image data set was established,and a simple convolutional neural network（CNN）was designed using the caffe framework to classify its closed state.Based on the classification results,combined with PERCLOS and yawning frequency,the driver’s fatigue factor is calculated every 30 seconds.The fatigue degree of the KSS subjective scale is divided,and the subjective rating of the driver is carried out every 10 minutes.The SVM classifier is used to predict the fatigue degree within 10 minutes according to its fatigue coefficient,and the accuracy of fatigue monitoring is verified.Finally,relevant experimental demonstrations were carried out in the simulated driving platform.Experimental results show that the proposed detection method can achieve high detection accuracy and is robust to changes in illumination and environment.