Research On 3D Modeling Of Human Face Based On Binocular Stereo Vision

With the development of Internet technology,the 3D modeling and visualization for current,3D facial modeling is a hot issue,from the current major enterprises face attendance machine to Alipay,mobile phone banking,brush face verification,are process modeling and recognition for face.Three dimensional modeling and reconstruction of human face can deal with the problem that the two-dimensional face is susceptible to illumination,expression and angle.In this paper,the 3D modeling method based on binocular stereo vision is studied.In this paper,by using two ordinary cameras to build a binocular stereo vision system,first use Matlab camera calibration tool to calibrate the camera,and obtain its internal and external parameters information.Then the dense matching algorithm is used to calibrate and match the calibrated camera.In the image preprocessing using bilateral filter method of dense matching before and after extraction of the face feature points by ASM method.According to these feature points constraint in the stereo matching process in the scope of the search,after comparing the fixed window of SAD,SSD,NCC method and adaptive weight method in this paper is used to calculate several face matching the cost of the matching effect,educed the using method of the best matching effect.In order to obtain more accurate disparity values,sub-pixel interpolation is used.And the left and right images are checked by consistency so that the left and right inconsistent points can be removed.Finally,the disparity map is obtained by median filtering,and the depth map is recovered successfully.This paper builds a visual platform system based on MFC,uses OPENCV to process input and output images,and uses OPENGL to face modeling.Finally,the Texas3 DFRD database is projected to the two images by using binocular camera imaging principle,and the algorithm is tested.The results show that the algorithm can reduce the matching time and reduce the false matching rate to a certain extent.In view of the needs of face modeling and reconstruction,this paper makes some exploration and tentative research on related algorithms and schemes,but there are still many shortcomings.In the following work,the training mechanism in ASM can be refined,and the accuracy of feature points can be greatly improved.In addition,the more accurate and reliable method can be further studied in the selection of adaptive weights.Facial 3D modeling andrecognition play an important role in more and more fields,and the research in this area will certainly meet the requirements of the modeling method.