Research On Driver Abnormal Behavior Detection Based On Machine Vision

Among the motor vehicle driving accidents in our country,the accidents caused by inattention due to driver fatigue driving and smoking,drinking and using mobile phones account for a large proportion.In the process of driving with a high degree of concentration,the driver is under a greater load physically and psychologically.In the state of fatigue,the driver’s slow movement and inattention caused by abnormal driving behaviors such as smoking and drinking water can easily lead to road traffic accidents.Therefore,the development of a system that can monitor driver fatigue and abnormal behavior in real time,timely detection and warning to stop fatigue driving and inattentive driving behavior is of great significance to reduce road accidents,ensure the personal safety of drivers and the safe operation of road traffic.In the current existing schemes,more attention is paid to rationally evaluating the vehicle state based on the overall state of the motor vehicle,which has the disadvantages of delay and neglect of human psychological attributes.In order to solve the above problems,this paper proposes a method based on Driver fatigue and abnormal behavior detection system based on face and behavior recognition.This paper mainly studies how to use machine vision and image recognition technology to accurately identify and detect the driver’s current driving state when the driver is in the state of fatigue driving and abnormal driving state of inattention,the basic principle of the detection algorithm and how to build the system The specific idea and specific method of the detection system.The basic face recognition uses the cascaded Adaboost face detector to detect the driver’s face as a whole,and uses the Dlib library and facial 68 key point capture technology to detect the driver’s eye area and The mouth area is accurately positioned and identified,and the real-time horizontal and vertical ratios of the eyes and the real-time opening of the mouth are calculated through the geometric coordinate relationship between key points,and the driver’s eye closure state and yawning state are further calculated through the PERCLOS algorithm.Therefore,based on these two basic states,it is determined whether the driver is fatigued and a real-time warning is given.At the same time,this paper uses the three object features of mobile phone,water cup,and cigarette as the target features of the driver’s three non-focused abnormal driving states,such as using a mobile phone,drinking water,and smoking.The abstract task of detecting the driver’s inattentive abnormal driving behavior state is embodied as the task of target detection for specific targets,so as to further realize the task of accurately identifying the inattentive abnormal driving state.At the algorithm level,this paper chooses the one-stage target detection network YOLOv5 model,and introduces the object key point attention mechanism,which greatly improves the detection ability of tiny targets and reduces the interference of redundant information in the detection background,the improved m AP value reaches 98.2%.The overall visual interface adopts the python pyside2 plug-in for comprehensive design,so that the driver fatigue state detection and the driver’s inattentive abnormal driving state detection are integrated in the same window,and the final design result can capture the driver’s facial dynamics in real time,and real-time detection and real-time warning of the driver’s fatigue driving state and abnormal behavior.