| Face recognition is one of the most active and challenging research fields of identification technology based on biological features. Compared with 2D face image,3D face model with crucial spatial depth data has more accurate and rich information, which may solve the problems in 2D face recognition caused by the variation of illumination, pose, expression and other factors, hence 3D face recognition theory, approach and technology become a hotspot in current studying. As 3D face recognition is based on face 3D data, the study of how to use computer technology to acquire full and accurate face 3D data is of great significance to3D face recognition.Under the guidance of Marr's computer vision theory, this paper makes a research on face 3D data acquisition approach based on binocular stereo vision.Firstly, camera model and existing camera calibration methods are introduced, on the basis of analysis and comparison these calibration methods, the camera calibration method based on OpenCV is adopted to work out the internal and external parameters of camera.Secondly, two images are obtained by two calibrated cameras filmed the same face from different perspective at the same time. A small amount of salt and pepper noise and the major Gaussian noise in two images first is removed by using the median filtering and the Gaussian filtering, and then histogram matching algorithm is used to balance brightness difference between two images.Finally, by studying advantages and disadvantages of the current stereo matching algorithm, the author puts forward an approach to stereo matching based on pair-wise alignment algorithm to achieve stereo matching of two images and obtain the corresponding match point of two images and the disparity map. According to the principle of binocular stereo vision, face 3D data can be obtained by using the disparity information of the corresponding matching points of two images as well as the internal and external parameters of the camera gained by calibrated camera. |
PDF Full Text Request