Design And Implementation Of Driving Behavior Recognition And Early Warning System Based On Deep Learning

Recently,with the fast economic development in our country and the construction of road traffic,the quantity of vehicles and drivers are remarkably increasing.Brings a great test to the traffic safety management system and emergency handling of accidents.Through the analysis of a large number of traffic accident cases,the main cause of vehicle traffic accidents is that the driver has many unrelated behaviors when driving,such as calling and checking the message.These behaviors distract the driver’s attention,causing the driver to be unable to pay attention to the surrounding traffic environment,resulting in traffic accidents.The occurrence of traffic accidents has brought huge economic losses and a large number of casualties to human society,In order to reduce the incidence of traffic accidents,it is necessary to give early warning to drivers’ unsafe driving behaviors.Most of the existing driving behavior early warning systems are cloud recognition models based on deep learning,which can well recognize the driving behavior of drivers.However,cloud recognition involves data transmission and information security during transmission.In addition,when the driver has unsafe driving behavior,needing to promptly remind the driver to pay attention to his personal driving behavior,which has higher requirements for the recognition efficiency of the cloud recognition model.Based on the above problems,the cloud recognition model cannot help the driver very well.Therefore,this paper uses a lightweight network model to solve the problem of recognition efficiency of the model.For data transmission and security during transmission,this paper constructs a local recognition model.At the same time,in order to remind users,design and implement a driving behavior early warning system.The main work of this paper is as follows:(1)Adopt a lightweight network model.By analyzing the existing network model,this paper selects the lightweight network model Mobilenet V3 as the driving behavior recognition model.Use the State Farm data set to train the model.Aiming at the problem of the efficiency of the model in the detection,this paper adopts the compression algorithm: model pruning and weight sharing.Experiments show that the parameters size of the compressed model are about 24% of the original model,which reduces the time required for model detection,and the recognition accuracy of the compressed model is 87.6%.After introducing channel shuffle to the convolutional layer,the recognition accuracy of the model is 89.3%,which meets the requirements of this article Task.(2)Adopt image brightness detection and image enhancement.In actual application scenarios,the pictures collected by the vehicle-mounted camera are more sensitive to light.Insufficient light leads to unclear pictures,which brings difficulties to the recognition of the model.In this paper,the Zero-DCE algorithm is used to enhance the image of the collected pictures.Since Zero-DCE consumes a lot of time in processing images,at the same time,the image captured by the camera is of better quality under sufficient light conditions,and the quality is poor under conditions of insufficient light.Therefore,there is no need to enhance all images collected.In order to distinguish whether the image brightness is normal,brightness detection is required.Determine whether the image is normal by judging the ratio of the average deviation of the grayscale image to the average deviation.(3)Design and implement a driving behavior early warning system.Following the idea of software engineering,this paper designs and implements a driving behavior early warning system.Aiming at the driver’s main use when driving in a car,and the scene is relatively fixed,this paper designs and implements a desktop application.The desktop application is deployed on the car terminal device.Based on the realization of the driving behavior recognition function,when the driver’s abnormal behavior is detected,the desktop application can issue an alarm sound to remind the user to drive safely.At the same time,the application visually displays each recognition task.For system administrator users,mainly traffic law enforcement personnel.This paper designs and implements the design and implements a Web system.Finally,this paper conducts a systematic test on the driving behavior recognition and early warning system.From the user level,the main functional modules of the system are tested,and they can respond to user operations and return correct information;from the server level,non-functional tests are conducted based on high concurrency,and the server can pass a certain level of concurrent access.