Research Of Human Faces Reconstruction On Binocular Stereo Vision

Human face3D reconstruction technology has been widely applied in more and more areas.Since the traditional three-dimensional reconstruction methods are complicated andimpractical, we designed a face3D reconstruction system which photographs two face imagesprojected by structured light in the stable scene through a binocular camera and to calculatesthe depth information through parallax.This paper analyzes and studies the human face3D reconstruction from the view ofbinocular stereo vision, and on the basis of this research, a practical face3D reconstructionsystem based on structured light is established. Our system consists of many parts likethree-dimensional calibration, image preprocessing, stereo matching, z-value counting, pointscurve displaying, and involves spatial geometry, projection geometry, computer graphics andimage processing and so on. As to the hardware, we use equipments like stereo cameras,projectors, and other homemade calibration chessboard and PC. As for the software, wedevelop a friendly and efficient high cohesion and low coupling three-dimensional model offace reconstruction system based on OpenCV and OpenGL.At the first phase, we designed a new three-dimensional calibration method. Throughcalibrating the camera, we got the internal and external parameters of the camera and thedistance between two lens so as to obtain the projection matrix. We used OpenCV tocomplete the three-dimensional calibration of the camera and make some adjustments to thelens distortion.During the second phase, we projected stripe-type structured light to the human face andget the face image with a structured light through the stereo camera. Then we preprocessedthe image such as filtering noise and graying it. We used the improved Canny algorithm toextract the structured light and optimize the edge information, and through a series of varyingand evolved condition detailing algorithm we got the image which can be accurately matched.In the third phase, we designed a stereo matching method based on two-step matchingmethod. Step one, we calculated the parallax roughly and established rice-shaped windows;then we did the least squares algorithm to pixels in the window in order to obtain thedeviation distance of the two images. Step two, we marked the stripes by Rank transformationand Census transformation to do an exact point match.At the fourth phase, according to the theory of binocular stereo vision, we calculated thedepth value through projection matrix and got a three-dimensional spatial coordinate of thosepoints. Then we completed the triangular gridding of points on the basis of…. And finallydisplayed the face3D model on the screen.