Research On Driver Abnormal Driving Behavior Based On Machine Vision

In recent years,Public Transport in China has developed rapidly.The safety of the bus is a matter of common concern of the whole society.In order to reduce traffic accidents caused due to the bus driver,to protect the personal safety of all passengers on the bus,the bus driver received increasing attention in the normative way of driving.For the current machine vision,abnormal driving safe driving behavior of the driver assistance systems are designed and implemented.They include distracted driving and fatigue driving.Safe driving behavior depends on whether the driver is in an abnormal behavior state.Machine vision approaches appears to bus drivers behavior recognition,helping determine whether the driver is focusing on the driving task.It has a high practical value and social significance.Therefore,this paper mainly discusses driving behavior of bus drivers for identifying abnormal driving behavior.In this paper,the contents are as follows:1.Within the bus driver driving process,the driver is the implementation of the driving behavior with his face and hands as key parts.We apply deep learning YOLOv3 and SSD detection algorithm on the object detection implementation.We get the basic state acts of the driver.2.By image processing,we find the positioned portion centroid coordinate of human face and hands.Then we propose SSD-based face-hand position network.The network expresses substantially state bus driver better,an accurate representation of the relationship between the hand portion and the face are built.3.Bus drivers answering the phone call,is a typical example of abnormal driving behavior.The applying of key frame extraction provides spatial features of the behavior.Similarly,the applying of multi-frame video extraction contains act behavior domain features.We get the bus driver hand trajectory while answering the call with respect to the face.Then we detect residence division,and time of stay in position around the face area.Phones are also detected to provide extra features.We build the model of phone calling answering behavior identification by combining the spatial-temporal domain characteristics and extra features.The experiments show that the mode 1 is feasible and effective.4.We build a common model of abnormal driving behavior of bus drivers,and verify the validity of the model with other abnormal driving behavior.The experiments show that the model is feasible and effective.Finally,we analyze the application of the model in bus scenes.