Research On Driver Abnormal Behavior Detection Based On Multi-information Key Points

With the development of social economy,the number of road vehicles increases gradually,and traffic accidents emerge one after another,a large part of which are caused by the dangerous operation of drivers.Driver abnormal behavior detection algorithm can effectively detect and evaluate driver’s behavior,in order to improve the safety of driver operation and reduce the occurrence of traffic accidents.In the process of driver abnormal behavior detection,the cluttered background and light in the car has always been a difficult problem that affects the detection accuracy,and the driver abnormal behavior is various,which may be accompanied by other items,which increases the difficulty of detection;limited by the fact that drivers are advised to wear masks in recent years,drivers’ behaviors such as drinking,smoking and eating have decreased,and the demand for mask wearing behavior detection has increased.In view of the above,the main research contents of this thesis are as follows:1)A behavior detection algorithm for driver operating mobile phone based on bone key points was proposed.In the process of vehicle driving,the real-time detection of the behavior of drivers operating mobile phones can provide effective help for reducing road traffic accidents.In view of the complex background in the car,which is prone to missed detection and false detection,a driver operating mobile phone behavior detection algorithm combined with bone key points is proposed.The algorithm constructs a parallel two-branch network structure,uses the Alpha Pose algorithm to recognize the human posture and makes a preliminary judgment on the driver’s behavior;uses the structured feature enhancement module to improve the YOLOv5 algorithm,and uses it to detect the mobile phone target in the driver’s hand area;integrates the human posture and hand detection information to comprehensively judge the driver’s behavior.The experimental results show that in the multi-directional and complex background environment,the m AP of driver operating mobile phone behavior detection is 97.7%.2)A driver’s one-handed steering wheel behavior detection algorithm based on attitude recognition was proposed.In order to solve the problem that the driver’s one-handed steering wheel behavior is diverse and the range of behavior is wide,a driver one-handed steering wheel behavior detection algorithm based on attitude recognition is proposed.The algorithm uses the Alpha Pose network model to detect the key points of the upper body bone of the driver,and then connects the key points to form human posture information,combined with the difficult sample mining technology to train the driving attitude model,and finally uses support vector machine to classify the target and output the detection results.The experimental results show that in the multi-directional and complex background environment,the m AP of driver operating the steering wheel with one hand is 96.03%.3)A driver mask wearing behavior detection algorithm based on cascaded convolution neural network was proposed.Driver wearing mask is an important part of safe and healthy travel.In order to solve the problem of low detection rate of driver mask wearing behavior in complex light environment in public places,a mask wearing behavior detection algorithm based on cascaded convolution neural network is proposed.The attention mechanism module is used to improve the MTCNN algorithm,and it is used to locate the driver’s face region frame;the results of the located face region are input into Mobile Net V2 for mask wear detection and classification,and finally output the detection results.The experimental results show that in the multi-directional and complex background environment,the mask detection rate and detection accuracy of this algorithm are93.88% and 91.75%,respectively.