Development Of A 3D Face Database Based On Binocular Vision

In recent years,face recognition technology has been widely used in finance,security,traffic,justice,medical and other fields.Compared with two-dimensional face image,three-dimensional face data is more robust to face attitude changes,scale changes and light source changes,and it is rich in more spatial structure information and is not easy to be copied.Therefore,three-dimensional face recognition technology is better than traditional two-dimensional face recognition technology Potential,gradually become the development trend of face recognition field.3D face database is an important data source of 3D face recognition and related disciplines.It has a variety of construction methods,and has important research value.Therefore,this paper aims to develop a small 3D face database by using the low-cost binocular stereo vision acquisition method.In this paper,the overall development process of 3D face database based on binocular vision is described.Firstly,the acquisition system is built.Based on the principle of binocular stereo vision,two CCD cameras of dh-sv1411 fm are used to build the binocular stereo acquisition system,and Zhang Zhengyou calibration method is selected to calibrate,and the important parameters of the two cameras are obtained.Secondly,the face data collection,a total of 30 samples were collected.Each sample collected 8 groups of neutral facial expression images with different facial posture.In the stereo matching stage,the existing stereo matching algorithm and key steps are analyzed in detail.For the weak texture region of human face,a cross scale cost aggregation model is adopted to divide multiple scale spaces.In different scale spaces,the cost aggregation is carried out by combining the guided filtering algorithm.Finally,an ideal disparity map is obtained and restored according to the principle of triangulation Face 3D data.In the stage of point cloud data processing,the point cloud data is preprocessed,and the complete and effective 3D data of face area is obtained.Using Delaunay triangulation algorithm to reconstruct the three-dimensional face model,taking thenose tip as the center point,establishing the corresponding relationship between the models,obtaining multiple depth maps of all samples,and storing all samples in a unified format.Finally,the experiment of 3D face recognition based on LBP is carried out,and the feasibility of constructing 3D face database using the method in this paper is verified by the experimental results.