Research On Driver Fatigue Algorithm Based On Machine Learning

In recent years,the social economy has developed rapidly,and the living conditions of ordinary people are getting better and better.People are no longer just satisfied with food and clothing,and begin to improve the quality of life.In the past,only a few businessmen owned cars,but now almost every household Everyone owns a small car.The car is no longer a luxury and has become an indispensable part of people's lives.The car is also a double-edged sword,which brings convenience as well as hidden dangers.The continuous increase of car ownership has also brought frequent occurrence of traffic accidents,which has greatly harmed people's lives and property safety.Will have a bad impact on the stable development of society.A large amount of data shows that one of the important reasons for traffic accidents is fatigue driving.In traffic accidents in China,the number of casualties caused by fatigue driving accounts for 20%of the total,and more than 40%of major accidents.In order to reduce traffic accidents due to fatigue driving,on the one hand,the fatigue driving management system can be improved to strengthen this aspect of management,on the other hand,scientific researchers can study the driver's physiological characteristics and use modern science and technology to drive.Remind staff when they are tired.Therefore,it is of great significance to study the fatigue detection of drivers.In recent years,the mainstream of fatigue driving research is an algorithm based on visual features.The detection algorithm has low detection cost,high speed,and strong robustness.However,in the actual environment,whether the driver wears glasses or sunglasses,changes in light will have a certain impact on the accuracy and real-time detection.Based on the research on driving fatigue detection algorithms at home and abroad,this paper proposes a driver fatigue detection algorithm based on machine learning.This method improves the accuracy,real-time and robustness of fatigue detection.The algorithm proposed in this paper combines multiple detection algorithms to better locate the face area by improving the full convolutional neural network,and then uses the improved Adaboost algorithm and Sobel edge detection operator to extract the eye area features,and then passes Convolutional neural network technology accurately detects the state of the human eye,and finally uses the PERCLOS fatigue determination method and blink frequency to determine the fatigue of the driver.The work done in this paper is divided into four steps:1)Preprocess the image.The images collected in a dark environment are supplemented with histogram equalization and reference white processing methods;median filtering and image graying are used to de-noise and image pre-processing.After this step,the recognizability of the device imaging is improved,and the detection accuracy is also improved.2)Face Detection.First replace the fully connected layer in the VGG network with a full convolutional layer,then replace the classification layer with two classifications,and finally perform face detection under this algorithm by multi-scale transforming the images to be detected and input them to the full volume In the product neural network,the corresponding probability matrix is obtained,and the face frame is obtained by the non-maximum suppression method.The experimental results show that the algorithm has higher accuracy,shorter detection time and better performance.3)Eye detection.Eyes are one of the important characteristics that reflect the state of the human body.Opening,closing,half-opening,an blinking frequencies can all well convey whether you are in a state of fatigue.This paper chooses to improve the Adaboost algorithm to detect human eyes,and to reduce the eye search area as much as possible.This algorithm can detect human eyes quickly and accurately.4)Fatigue identification.Based on the driver's eye location model,this paper combines the P80 criterion and blink frequency in the PERCLOS criterion to detect fatigue of the driver,which improves the robustness of the algorithm.Various fatigue detection algorithms in this article are completed in the PyCharm environment,using the Python language and OpenCV machine vision library.Experimental results show that the algorithm can effectively detect the fatigue state of the driver,and the detection accuracy and real-time performance are also improved.