Research And Implementation Of Vehicle-mounted Dangerous Driving Behavior Alarm System

With the development of social economy,automobiles have become one of the main means of transportation for people’s convenient travel.The popularity of automobiles not only improves travel efficiency,but also brings traffic safety problems.How to improve driving safety has become a current research hotspot.The driver’s dangerous driving behavior is one of the main causes of traffic accidents.The vehicle-mounted dangerous driving behavior alarm system studied in this thesis aims to identify the driver’s dangerous driving behavior in real time through computer vision algorithms and use voice alarm prompts to improve driving safety.In this thesis,a driving behavior dataset is created according to the actual usage scenarios of the system and the driving habits of Chinese people.The data set includes nine types of driving behavior data,including normal driving,side view,drinking water,smoking,operating the central control,playing with mobile phones,holding things sideways,arranging appearance,and answering the phone.The training set and the test set are divided according to the collecting personnel.Compared with existing public datasets,this dataset has more advantages in terms of data volume and data richness.This thesis proposes a driving behavior recognition algorithm that utilizes the steering wheel to locate key areas.The algorithm firstly locates the steering wheel position in the car through the object detection algorithm,then uses the steering wheel position information to cut out the key areas,and finally classifies the key areas through the classification algorithm.Since the steering wheel position does not change in the same video stream,the position information can be reused,successfully transforming the object detection task into a simpler classification task.In order to deploy the driving behavior recognition algorithm directly in the vehiclemounted device,a lightweight neural network is used to implement the classification algorithm.According to the characteristics of the self-made data set,this thesis designs a lightweight hourglass module to improve the fully connected part of the Mobile Netv2 network,which improves the recognition accuracy while reducing the amount of network parameters.This thesis proposes a driving behavior risk assessment algorithm,which combines the driving behavior classification results,behavior duration and driving speed to accurately evaluate the current driving risk and improve the accuracy of voice prompts and violation records.This thesis uses the self-made data set to train the improved Mobile Netv2 network,and realizes the driving behavior recognition algorithm and risk assessment algorithm.Finally,the above algorithms are applied to the vehicle-mounted dangerous driving behavior alarm system.The vehicle-mounted end of the system is deployed in the vehiclemounted device,which can identify the driver’s dangerous driving behavior in real time and give a voice alarm prompt to improve driving safety.The platform end of the system is deployed in the server to collect driver violation records saved on the vehicle-mounted end of the system,which is convenient for users to analyze and manage driver violation records.