Research And System Implementation Of Dangerous Driving Behavior Detection Method

Fatigue driving and distracted driving are the main causes of road traffic accidents in dangerous driving behaviors.The detection and timely warning of dangerous driving behavior can effectively reduce the incidence of road traffic accidents and ensure the safety of people’s lives and property.Therefore,the research on the detection method of dangerous driving behavior has become a major research hotspot in recent years,and achieved good results.However,on the edge devices with limited resources,dangerous driving behavior detection methods face the challenges of weak adaptability to illumination environment and low accuracy.Aiming at the above challenges,the research and system design of efficient dangerous driving behavior detection method with strong adaptability to different illumination environments are carried out in this thesis.The existing detection methods of dangerous driving behavior are deeply investigated,and the advantages and disadvantages of the detection methods based on physiological characteristics,behavioral characteristics and computer vision are analyzed in this thesis.A set of dangerous driving behavior detection method and dangerous driving behavior detection system scheme with strong illumination environment adaptability and low computational complexity are proposed in this thesis.On this basis,the video data of driver behavior under different illumination environments are collected,and the fatigue and distraction behavior datasets are produced in this thesis.The neural network model constructed in this thesis is compared with other methods on public datasets.To provide better validation of the performance of the dangerous driving behavior detection method proposed in this thesis,the dangerous driving behavior detection system is designed based on NVIDIA Jetson TX2 embedded platform and infrared light supplement camera,and the function and performance of the system are tested under the complex illumination environment in the car.The innovations of this thesis are as follows: 1)An illumination environment recognition method based on OTSU and fuzzy C-means clustering algorithm is proposed.The combination of illumination environment recognition method and lightweight convolutional neural network model can effectively enhance the illumination environment adaptability of fatigue driving detection algorithm.2)An inverted residual attention module is designed,and a lightweight convolution neural network model based on attention mechanism is constructed.The combination of the subtract mean filtering algorithm and the constructed network model can effectively reduce the computational complexity of the distracted driving detection algorithm while ensuring the accuracy.After testing,the detection accuracy of the dangerous driving behavior detection system designed in this thesis reaches more than 95% under the complex illumination environment in the car.At the same time,it has the functions of real-time warning of dangerous driving behavior and evidence preservation of dangerous driving behavior.