Research On Intelligent Monitoring Method Of Drivers’ Bad Driving Behavior Based On Deep Learning

In recent years,with the rapid growth of the number of cars in our country,drivers have violated the "Road Traffic Safety Law" during driving,making and answering calls,smoking and other bad driving behaviors,as well as the traffic safety hazards and traffic hazards caused by distracted driving.The problem of accidents is still relatively prominent,and has attracted widespread attention from all walks of life.It is urgent to adopt more effective means to strengthen supervision and effective governance.For a long time,the monitoring of drivers’ driving behavior has mostly been based on the monitoring of external hardware equipment.There are problems such as high system investment costs,large resource consumption,and difficulty in popularizing the use of the entire road network.In addition,the driving environment is complex and there are many uncontrollable factors,the problem of false detection and missed detection has not been well resolved.In recent years,with the rapid development of computer vision technology,especially the rapid rise of artificial intelligence technology represented by convolutional neural network,automatic driver behavior monitoring technology based on artificial intelligence and vehicle video has gradually become the focus of research in this field,and is considered to be one of the most effective ways to solve the problem of driver’s "bad driving" behavior monitoring on a larger scale.Based on the above background,this thesis takes the vehicle-mounted monocular camera as the traffic scene video acquisition scheme,combines the characteristics of the actual driving room scene,comprehensively uses artificial intelligence technology,machine vision and pattern recognition technology,and takes the monitoring of drivers’ behaviors such as illegal dialing and answering and smoking during driving as the case study.An in-depth study is carried out on the automatic detection technology of drivers’ bad driving behavior.The main research work and research results are as follows:(1)In view of the fact that bad driving behaviors such as hand-held phone calls and smoking will cause characteristic changes in the video image of the driver’s head area,in order to achieve fast and accurate dynamic monitoring of bad driving behaviors,a model based on YOLOv5s-Siam RPN++ is proposed Automatic calibration and tracking method of driver’s head region of interest.This method first uses the good target detection performance of the YOLOv5 s model to identify the driver’s head area,and then inputs the recognition result as a target image to the Siam RPN++ tracking network for real-time tracking.On this basis,the optimal position and size mapping relationship between the head area and the area where bad driving behavior can be determined is established by cluster analysis method,and then the automatic calibration and real-time tracking of the driver’s head area of interest are realized.The experimental results show that this method outperforms traditional algorithms in tracking accuracy of targets,and the ROI of the head is set reasonably.(2)In order to improve the recognition accuracy of small objects such as phones and cigarettes in the area of interest of the driver’s head as the main basis for determining bad driving behavior,the YOLOv5 s model algorithm is used in actual application scenarios due to hand occlusion and other reasons.For the problem of missed detection and high false detection rate of small targets such as phones,a specific small target related to bad driving behavior based on YOLOv5 s improvement is proposed by adding a self-attention module to the YOLOv5 s backbone network,optimizing loss functions and activation functions.detection algorithm.The experimental results show that the improved model has a good performance in improving the detection accuracy and recall rate of specific objects,and can monitor the driver’s hand-held phone behavior in real time during driving.(3)In view of the judgment of the driver’s smoking behavior,it is not only related to whether there is a cigarette target in the area of interest,but also related to whether there is smoking smoke.For this reason,aiming at the formation and dissipation of smoking smoke in the cab,a further research is carried out to propose a recognition algorithm for smoking smoke in the cab based on HSV color space.The experimental results show that this method can improve the detection accuracy of the smoke caused by the driver’s smoking.With good performance,combined with the recognition results of small objects such as cigarettes in the area of interest of the driver’s head,it can automatically monitor the driver’s smoking behavior.The model algorithms proposed in this thesis are all implemented in the Python language,and are verified by multiple sets of experiments on the experimental data.Based on all the experimental results,the deep learning-based automatic monitoring method for driver’s bad driving behavior proposed in this thesis can better meet the technical requirements in real application scenarios,and has the value of application and promotion.