Three-dimensional Reconstruction Of Human Head Using Multi-eye Vision

The purpose of stereo vision is to obtain three-dimensional information from images of several different position in the same scene.Stereo vision is mainly composed of five parts:feature extraction,stereo matching,camera calibration,3D information recovery and post-processing.This paper focus on the computer stereo vision system,aiming at intersection testing of the medical headgear,intersection matching,camera calibration,extract the 3D information and triangulation of point cloud.In feature point detection and matching,this paper uses Shi-Tomasi algorithm to detect intersection points on medical headgear.Also,an improved method is proposed,which uses this algorithm to detect intersections at different scales and to re-detect and further locate intersections.Due to the similarities on the medical headgear,this article uses ten colors to color the headgear.Cross-point matching is performed by the cross point pair,polar-line constraint and the intersection color information.Finally,the RANSAC algorithm is used to remove the mismatched points in the matching process,so as to improve the matching effect of the cross points.In the part of camera calibration,a method of Zhengyou Zhang which is between the traditional calibration and the self-calibration to obtain the inner and outer parameters of the binocular camera is adopted in this paper.Four cameras are used to take photos of the head.The primary camera is calibrated with three other cameras,the obtained internal and external parameters provide convenience for calculating the coordinate information of 3D points.In the 3D information acquisition,the basic matrix is calculated by matching points on the medical headdress,and then the intrinsic matrix is calculated according to the internal parameters of the camera.Finally,the projection matrix of the camera is acquired according to the external parameters of the camera to calculate the 3D coordinates of the matching intersection points.Point cloud integration will be around the three-dimensional points of the head were unified into the respective main camera coordinate system,then the left and right parts of the head cloud are unified into a same three-dimensional coordinate system.Medical headdress can be approximated as a triangular grid,it can triangulate the acquired point cloud.In this paper,we compared the conventional Delaunay triangulation method and the Crust method.The Crust method can well triangulate the three-dimensional points.