Research On Driver Fatigue Detection Based On Facial Features

Fatigue driving is one of the important causes of traffic accidents.Therefore,how to effectively perform driver fatigue detection has become a hot research field.In this paper,a detection algorithm based on face key point location is proposed for the driver's fatigue driving state,which can effectively detect the driver's dangerous driving state.The specific work is as follows:1.Optimize the image captured by the camera by median filtering,histogram equalization,etc.Attenuate the effects of illumination and noise on the image.MLP regression calibration network is added after classic Adaboost face detection classifier to obtain a more accurate and stable face regression frame,which provides accurate face position for face key point positioning.2.On the basis of face detection,the key points of the face are located by improved TCDCN network.Build a branching strategy for different tasks according to the convergence difficulty of different subtasks.Other than this,the early stop strategy of TCDCN was improved to ensure that the key point regression task converges faster to get 9 face key points.3.Research on the fatigue detection method based on machine vision.According to the PERCLOS fatigue judgment criterion,the state feature judgment method related to driver fatigue detection is proposed.According to the nine key points of the face,the position of the human eye is extracted,and a two-category neural network is trained to judge the closed state of the eye.At the same time,according to the positional relationship of the nine points,the orientation of the face is judged,and the head state of the driver is given.Whether the driver is fatigued or not is judged by the state of the head and the eye.4.In this paper,the face detection algorithm and key point positioning algorithm are tested and compared experimentally.Aiming at the face detection algorithm,this paper compares the new Adaboost Face detection algorithm with the classical Adaboost face detection algorithm on mixed dataset,and the results show that the accuracy and IOU of this algorithm are better than the original Adaboost algorithm.Aiming at the key point detection algorithm,this paper discusses the influence of optimizer,loss function and network structure on the original TCDCN network,and the experimental results show that the improved TCDCN network performance in 9 face key point datasets is better than the original TCDCN network effect,the model converges quickly and the feature point location is more stable.