Font Size: a A A

Research On Face Pose Estimation Andrecognition System Based On Feature Point

Posted on:2018-03-02Degree:MasterType:Thesis
Country:ChinaCandidate:P C DuanFull Text:PDF
GTID:2348330518998543Subject:Engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of computer technology,human no longer meet the interactive mode with computer with the help of external hardware equipment,but rather to want the computer to become a machine that can communicate with oneself in natural language,can communicate with people more convenient,more intelligent.At this time,the face will play an important role.So the need for in-depth study of the face.Moreover,we will use the main information of face to the virtual glasses test system in the online glasses shop.Users can experience the effect of wearing glasses keep indoors.Facial feature points are the key to the face recognition,3D pose estimation and virtual eyewear trial,the accuracy of feature point positioning to a large extent affect the performance of these applications.Based on the face detection,this paper extracts the local binary features of face images and uses the idea of random forest to return each feature point to the target position.In this paper,the feature point of the key position of the face is studied deeply,and it is applied to the face pose estimation,face recognition and virtual eyewear test system,and the detailed description of these parts is done.Finally,we also improve creatively the relevant algorithm.The main work of the paper is as follows:1)Describing a feature extraction method which is based on the region of extracted feature.In this paper,the accuracy of the facial feature points is improved by changing the number of face images in the training database,the number of trees in the random forest and the number of trees.Experiments show that the algorithm has strong robustness in complex environment.It is found that the traditional feature extraction method has large feature dimension,which leads to a large amount of computation when the similarity measure is made.Based on the feature points,a method of local Gabor transform in the neighborhood of face feature points is proposed.Experimental results show thatthe method not only can effectively extract the local feature information of the face,but also can greatly reduce the dimension of the feature and speed up the face Identify the speed.The experimental results show the effectiveness of the method.2)A method of three-dimensional pose estimation is proposed.Based on the three-dimensional posture of the human face,it is found that the traditional face pose estimation method can only achieve the angle estimation on four degrees of freedom,and it is difficult to simulate the actual situation of face transformation.And proposes a three-dimensional attitude estimation method for human face.This method can estimate the angle of 3D face pose transformation by using the affine transformation between two-dimensional coordinates and the corresponding three-dimensional coordinates on the basis of precise positioning to the feature points of face key position,and can realize 6 degrees of freedom,30 degrees within the range of accurate estimates.And this paper realizes a virtual eyewear trial system based on facial feature point location.The system can not only real-time estimate the rotation of the face,and can track the location of the user's eyes in real time.
Keywords/Search Tags:facial feature points, three-dimensional pose estimation, virtual eyewear trial, face recognition
PDF Full Text Request
Related items