Research On Driver Attention Monitoring And Warning System Based On Lightweight Network

Traffic safety is the focus of social concern,and the dangerous driving behavior of drivers is one of the main causes of traffic accidents.The current traffic supervision equipment still cannot detect the initial dangerous driving behavior of drivers,and the attention monitoring early warning system can send out early warning signals at the early stage of dangerous driving to reduce the traffic accidents caused by inattentive drivers.In recent years,deep learning methods have developed rapidly and played an important role in the field of computer vision,and their powerful image feature extraction capability is often used to solve complex image problems.In this study,a multi-task learning method is designed using a lightweight backbone network to implement a mobile driver attention monitoring and warning system by detecting the driver’s head posture and facial features.The research method used in this thesis has the advantages of speed and low cost compared to previous studies.The main research elements of this thesis are as follows:(1)A multi-task learning method for face pose estimation and face alignment is implemented using a lightweight backbone network,which improves model accuracy and robustness.The model structure and evaluation function are redesigned to reduce the model parameters and meet the real-time and accuracy requirements of the system for deep learning algorithms while ensuring approximate accuracy with the mainstream algorithms.(2)Two monitoring methods with different attention characteristics are proposed.One is to design a driver gaze area detection method for the driver’s behavior of having vision out of the safe driving area.The method guides the driver to customize parameters through the system language and uses the interconversion of the world coordinate system and the camera coordinate system to complete the gaze area detection.Second,for the driver’s inattentive behavior with facial features such as fatigue and low gaze degree,the fatigue degree is monitored using eye features and mouth features,and the gaze degree of the driver is analyzed using pupil features.The two attentional feature monitoring methods can be customized with parameters and algorithm weights according to different drivers and vehicles,improving the practicality of the system.(3)A driver attention monitoring and warning system running on mobile was designed and developed.The system uses the mobile device’s camera,sensors,and speakers to work in concert.Multiple algorithms proposed in this study are used to determine the driver’s attention state and combine with vehicle motion characteristics to determine whether a real-time warning is needed.Unlike traditional fixed camera systems,the mobile device equipped with this system can be placed anywhere in front of the driver,and the device sends an early warning signal when the driver shows characteristics of reduced concentration.The system uses only the computing power of the mobile device to monitor the driver’s attention status in real-time.