Fatigue Driving Detection Based On Deep Learning

With the rapid progress of social development,people's living standards are constantly improving,because of the widespread use of cars resulting in frequent traffic accidents,it is counted that 40% of traffic accidents are caused by driver fatigue driving.In response to this problem,researchers have proposed many fatigue detection methods,including physiological information based method,vehicle behavior based method,and vision based method.Among them,the driving vision based detection method has the advantages of non-contact,high accuracy and easy implementation.So in this paper,the analysis on driver's facial features is used to determine the fatigue state.Firstly,this paper studies the detection methods of fatigue driving at home and abroad,analyzes the advantages and disadvantages of the existing algorithms,and puts forward a fatigue driving detection method based on deep learning.In this method,the vehicle camera is utilized to identify the state of the driver's eyes and mouth,and determines whether the driver is tired by the way of multi-index fusion.Its main tasks are as follows:(1)Face detection and tracking.First,the common face recognition algorithms is studied and analyzed,and finally the Multi-task Cascaded Convolutional Networks(MTCNN)is selected to detect the driver's facial features.The MTCNN network fully considers the potential relationship between face recognition and face key points,and can detect both of them simultaneously.After obtaining the key points,the eye and mouth regions are intercepted according to the distribution of face organs.Taking into account the problem that the detection accuracy of the MTCNN algorithm decreases when the face is deflected,occluded and sunglasses worn,a new method combining the MTCNN algorithm with the improved Discriminant Scale Space Tracking(DSST)algorithm is proposed,which solves the problem of face detection and tracking well.(2)Eyes and mouth states judgment.After obtaining the key points,the eyes and mouth regions are intercepted and the state of the eyes and mouth is recognized in combination with the distribution of the facial organs.A judgment algorithm based on Mobile Net V2 lightweight network model is adopted here,which can detect the open and closed state of eyes and mouth.On the premise of ensuring the detection accuracy,using lightweight neural network can reduce the complexity of the model,speed up the operation,and better meet the real-time requirements of fatigue driving detection.(3)Multi-index integration based fatigue judgment.By collecting the fatigue indexes such as the value of Percentage of Eyelid Closure Over the Pupil Over Time(PERCLOS),blink frequency,eye closing time,yawning frequency and so on,the multi-index fusion method is used to judge whether the driver is tired or not.While a certain index is failure in the detection process,the rest of indexes can still make a correct and timely judgment,gives the prompt warning to the driver,avoids underreporting.The simulative experiments show that the new algorithm can make a real-time and accurate judgment on the fatigue state of the driver,and make an early warning to remind the driver to adjust the state or stop and rest to avoid the adverse consequences.