Development Of Driving Behavior Analysis System Based On Machine Vision And Vehicle Driving Data Fusion

With the fast development of economy and the traffic roads was continuous constructed in our country in the past few years,the total number of vehicles and drivers have both increased significantly.Development of transportation not only brings convenience to our life,but also brings hidden danger.At the same time,it is a great challenge for road traffic safety system and emergency handling of traffic accidents.With the complexity of road traffic,the problems of driving behavior are also increasing gradually.The occurrence of traffic accidents is largely caused by abnormal driving behavior through the analysis of many traffic accident cases.There are a lot of redundant behaviors,such as drinking water,smoking,using mobile phones,over speed,driving fatigue and so on.These behaviors will greatly reduce the concentration of driver,make driver unable to pay attention to their own vehicle status and the surrounding environment,resulting in the occurrence of traffic accidents.This paper based on the fusion of computer vision data and vehicle driving data,studies the detection and recognition of driving behavior.The specific work is as follows:Firstly,according to JT/T808 communication protocol,the vehicle driving data is collected by using GPS terminal and wireless 4G/5G network,and uploaded to the cloud server for remote storage.On the cloud server,based on SOCKET communication,DOTNET technology and C# language are used to develop the data receiving service.The service analyzes and processes the data messages which uploaded by the vehicle terminal,and save them to the database for behavior analysis.Secondly,PERCLOS algorithm is used to detect and analyze fatigue behavior.Face detection is carried out based on Dlib,68 feature points of face are extracted,feature points of eyes and mouth are located,the eye aspect ratio EAR and the mouth aspect ratio MAR are calculated.PERCLOS is used to identify the fatigue of driver and judge the current state of driver comprehensively.Thirdly,the CNN network of YOLO is used to detect behaviors such as smoking,drinking water and using mobile phones,and analyze the confidence.The training module provided by YOLO is used to conduct training and generate model files.When YOLO is used for target detection,the model file is called to recognize the image,and the identification frame and confidence are displayed on the image frame.Finally,The SQL Server database on the cloud server is used to store vehicle driving data and driver image data.The vehicle driving data and driver video data are integrated and analyzed to find out the key indicators which reflect various abnormal driving behaviors,and the number of abnormal driving behaviors is counted.The entropy weight method and analytic hierarchy process are applied to compute the weight value of driver’s behavior score index.On this basis,the driving behavior recognition model is constructed,and the driving behavior scoring system is designed and implemented.