Research On Driver Behavior Analysis Based On Monocular Camera

In 2020,the number of cars in China has reached 367 million,which is increasing year by year.The frequent occurrence of traffic accidents is greatly increased due to the extensive use of cars.Most of these accidents are caused by the uncivilized and unsafe driving of drivers.Standardized driving behavior is not only the fundamental guarantee of driving safety,but also the precondition of avoiding traffic congestion,and also the precondition of property safety of pedestrians.Although the traditional method of detecting dangerous driving by facial features has been applied in practice,it is easily affected by many factors such as illumination and angle,which leads to low robustness and low accuracy of detection results.Therefore,this paper uses machine vision technology to estimate and recognize the driver’s behavior in real time,and its research has important theoretical significance and application value.In this thesis,the driver’s skeleton data is obtained by means of attitude estimation,and based on this,the driver’s behavior identity is realized.Firstly,the driver’s attitude estimation method is studied,and its implementation principle and usage method are expounded.The implementation principle and process of attitude estimation based on bottom-up Open Pose model are introduced.The driver skeleton sequence is obtained and stored in JSON format in a folder.Then,the obtained skeleton model is optimized by the simulated annealing fusion median filter optimization algorithm,and the optimized skeleton sequence of the model is Shi Kongtu.By comparing several classification methods of behavior recognition,a behavior recognition method based on spatio-temporal graph convolution neural network algorithm to train self-made driver database is proposed.Finally,based on the above research contents,the database,model training and algorithm program of the software system are developed and designed by using python language in Visual Studio development platform,and finally the driver behavior identity is successfully constructed.The system is divided into four modules:attitude estimation,training data,data analysis and behavior recognition.Each module realizes different functions,and finally completes the driver’s behavior recognition and analyzes the experimental results.Experimental results show that the driver behavior analysis system based on monocular camera can accurately identify the four driving behaviors of drivers: normal,one-handed,answering and calling,and fatigue driving.Applying the behavior identity to remind and warn drivers of dangerous driving behaviors of motor vehicles can effectively reduce the occurrence of traffic accidents,reduce the loss of public facilities and ensure the safety of people’s lives and property.