Research On Head Pose Estimation Algorithm Based On 3D Model Reconstruction

As an important research topic in the field of computer vision,head pose estimation has received extensive attention from researchers in recent years.At present,head pose estimation is widely used in many important fields such as safe driving,intelligent monitoring,virtual reality,and face recognition.At present,in the mainstream method,head pose estimation is regarded as a supervised classification or regression problem.The result depends largely on the accuracy of ground truth labels of training data,but in actual scenarios,it is difficult to obtain accurate labels of angles.The collection scenarios and protocol standards of existing public data sets are different,which will lead to different distribution characteristics of labels of these data sets.In order to solve the dependence of head pose estimation on data sets with different label distribution characteristics,this paper proposes a head pose estimation method based on 3D model reconstruction.This method is based on geometric mapping of feature points of 2D images and 3D models,and uses neural networks for mathematical calculations and optimization,which can reduce the dependence of head pose estimation on data labels to some extent.This paper mainly studies the head pose estimation algorithm based on 3D model reconstruction,including 3D model reconstruction,face feature point detection and head pose estimation.In 3D model reconstruction,the algorithm of 3D model reconstruction based on deep neural network is selected to return 3D deformation model parameters from 2D RGB image,and the training data set required by the experiment is synthesized by this method;in face feature point detection,considering the accuracy of face feature point detection,the feature point detection method based on face alignment network is selected;in head pose estimation In addition,neural network is used to replace the traditional mathematical iterative optimization process,and the loss of fitting 2D image feature points and 3D feature points is proposed.Based on the idea of 3D model reconstruction,the method proposed in this paper combines fitting of feature points with calculation of neural network,and has better performance in accuracy and stability than existing methods.Extensive experiments have proved the effectiveness of this method,so this method is suitable for most application scenarios of head pose estimation.