Research Of 3D Face Reconstruction Based On Binocular Vision

In recent years,with the development of information technology,and the rising of people living standard,in the security monitoring,game development,film and television production,human-computer interaction,etc,to obtain more detailed and realistic 3D face model has become strong demand.In order to overcome the 3D face reconstruction technology based on traditional laser scanning or structured light scanning of high cost,the equipment complex,the long sampling time,defects such as object attitude must be fixed.A growing number of researchers to study 3D face reconstruction method beased on binocular vision technology.This method has advantages that system hardware configuration is simple,the cost is low,the sampling speed and for object more friendly,the modeling precision is not sensitive to the face pose changes,etc.In this paper,the method for the system elaboration,mainly research the most core stereo matching steps in the process of reconstruction.The characteristics and innovations of this paper are listed as follows:(1)In stereo matching,in order to improve the matching precision of the face low texture area,this paper analyzes the causes of the matching accuracy decline in low grain area,propose the stereo matching algorithm based on epipolar distance transform.Because the epipolar distance transform has affine invariability,and the ability to turn face image on the gray level information to less affected by illumination,such as geometry information,to overcome the defects that precision declined caused by the low image texture area whose gray values is similar in most pixels and statistical features are uneasy to distinguish.(2)Stereo matching algorithm based on epipolar distance transform in the land of the texture region has a good matching effect,but near the edge of the areas or texture rich areas of the region,the robustness of the algorithm is not strong,the matching accuracy is not high.For this defect,this paper analyzes the performance of different texture area after the epipolar distance transformation,,the original epipolar distance transformation is improved,and put forward a biepipolar line length of epipolar distance transform,make it able to handle for this low texture region and high texture region hybrid complex situation,and improve the original epipolar distance transformation's robustness in high texture regions.(3)In this paper,in order to test the effectiveness of the algorithm we built a image acquisition platform,which can automatically complete the camera calibration,the acquisition of image to eliminating distortion correction,then use the parameters obtained from the calibration to the epipolar line correction,after the epipolar distance transform to the image which is corrected,the resulting images can be directly used for validation of matching algorithm.Experiments show that the proposed algorithm in the the human face image pair,compared with other traditional algorithm,has better accuracy and robustness,and compared to the popular stereo matching algorithms,it also can get similar performance when dealing with general scenario.