| With the continuous development of the economy and society,and the continuous progress in the field of computer vision and artificial intelligence,head pose estimation has received more and more attention in various fields.The robust head pose estimation algorithm has important application value in human-computer interaction,virtual reality,social activity analysis and driver assistance system.Head pose estimation is usually affected by factors such as face pose,expression change and occlusion,and inaccurate feature extraction,which leads to the low recognition rate.This paper proposes a head pose estimation method based on local binary features.The main research work of this paper is as follows:Firstly,in the face detection part,the face detection algorithm based on Haar feature and Adaboost cascade classifier is used to perform pre-processing of face image,accurately locate the face position in the image,and reduce the calculation amount for subsequent work;In the facial landmark location part,in order to further improve the location accuracy of the key facial feature points,the facial key point localization algorithm based on local binary features is used to extract the local binary features by using pixel difference feature.By training the random forest to gradually return to the real shape based on the current shape,the precise location of the key facial feature points has been realized.At the same time,the local binary features are highly sparse,which makes the running speed of the algorithm greatly improved.Secondly,in the head pose estimation part,in the head pose estimation scene of static image,because BP neural network has the characteristics that can realizes classification of any complex nonlinear map,proposed a head pose estimation method based on local binary features and BP neural network,input the position coordinate feature of the key facial point into BP neural network,and performs the training of the classifier to establish a mapping relationship between target output results of different head poses and input vector,so that realized the accurate estimation of the head pose of the static image.The POSIT algorithm is used to solve the pose parameters,only less input information is needed to obtain accurate head pose parameters through continuous iteration,which ensures the accuracy of head pose estimation and the real-time estimation requirement is realized at the same time.Finally,a set of fatigue driving detection system is designed.In order to solve the problem of misjudgment and low accuracy of a single fatigue driving judgment standard,a multi-index comprehensive fatigue driving judgment method based on head posture and eye closure degree is proposed,the comprehensive head posture parameter and the degree of closed eyes reflected by the relative distance between the feature points of the two eyes collectively judge whether the driver has a fatigue driving state,and effectively improve the accuracy of the fatigue driving judgment.Through effective simulation experiments,it is verified that the method solves the problem of inaccurate location of key facial feature points,the accuracy of head pose estimation is improved,and the accurate detection of fatigue driving state can be realized at the same time,it has a high running speed in the simulation system,the requirements of the accuracy and real-time of fatigue driving detection were met. |
