【作者】 吴果强；

【导师】 吴禄慎；

【作者基本信息】 南昌大学 ， 精密仪器及机械， 2010， 硕士

【摘要】 随着计算机图形学和计算机智能化的不断发展,三维人脸模型重构在安全认证、影视动漫、游戏娱乐、医学等许多行业发挥着极其重要的作用,因此,重构出具有真实感的三维人脸模型是三维人脸检测技术的研究热点,也是三维检测与逆向工程的重点研究内容之一。本文主要是针对三维人脸重构与反求技术进行研究,主要包括三维人脸点云数据获取、三维人脸点云数据重构处理和三维人脸CAD模型逆向加工三部分。1、三维人脸点云数据获取。介绍了基于结构光投影光栅相位法相调制原理,分析了传统的一维去包裹算法并推广到二维去包裹算法,比较了投影光栅相位法解相的相移法PSP和傅立叶变换法FTP两种算法的优缺点,最终选择基于二维傅立叶变换法的人体面部数字化扫描系统来获取精确的三维人脸点云数据。2、三维人脸点云数据重构处理。介绍了旋动理论,阐述了旋转矩阵R和平移矩阵T详细求解过程,分析了基于贴标志点的粗拼接算法和基于最小二乘法和奇异值分解法相结合的高精度自动拼接算法；提出了利用凸台形标定块实现两片人脸点云数据自动拼接。三维人脸点云数据经粗、精拼接处理后,生成具有真实感的三维人脸纹理图,然后重构出逼真的三维人脸NURBS曲面,最后形成三维人脸CAD模型。3、三维人脸CAD模型逆向加工。针对人脸复杂曲面,采用逆向反求技术来加工三维人脸实体模型,具体拟定了两种逆向加工方案。方案一是基于CATIA三维设计软件的数控铣床三维人脸模型加工,主要是进行了三维人脸仿真加工；方案二是基于MEM-350快速成形系统三维人脸模型加工,主要是基于ABS材料的三维人脸实物模型加工。本文的研究工作成果对建立一个具有鲁棒性的三维人脸识别系统研究以及相关学科的发展发挥着积极的推动作用。更多还原

【Abstract】 With the development of Computer Graphics and Computer intelligence, the reconstruction of 3D-face model plays an extremely important role in many industries, such as the security certification, the film and television, the anime, the games and entertainment, the medical science and so on. Therefore, to reconstruct a realistic 3D-face model is not only the hotspot of 3D-face detection technology,but also is one of the research emphasizes of 3D detecting and reverse engineering..Depending on the research to 3D-face reconstruction and reverse engineering, the paper contains three parts, which are the acquisition of 3D-face point cloud data, the processing of the point cloud data and the reverse processing of 3D-face CAD model.1. The acquisition of 3D-face point cloud data. Firstly, modulation principle of projection fringe phase method based on structured light was introduced, with the analysis to the two-dimensional unwrapping algorithm extended from the traditional one-dimensional one. Then two algorithms to solute phase which are Phase Shift method (PSP) and Fourier Transform method (FTP) were compared with each other. Finally, the facial digital scanning system based on 2D Fourier Transform method was adopted to obtain the accurate point cloud data.2. The reconstruct processing of 3D-face point cloud data. The gyration theory was introduced, and the detailed solution process to rotation matrix R and translation matrix T were explained by the paper. Then with the analysis to the rough stitching algorithm based on posted signalized point and the high-precision automatic stitching algorithm based on the combination of the least square method and the unit quaternions, in which Boss-block calibration landmarks achieving the automatic matching of two faces point cloud was proposed. After that, faces point cloud data mosaic.3D point cloud data of face coarse and fine stitching after treatment, the realistic 3D-face texture image was generated, and the realistic 3D-face NURBS surfaces was reconstructed, finally a 3D-face CAD model was formed.3. The reverse processing of 3D-face CAD model. According to the complex face surfaces, the reverse engineering technology was adopted to process 3D-face solid model, and two reverse processing programs were proposed in the paper. One is using numerically controlled fraise machine based on CATLA software to process the model, which is mainly advanced on the simulation processing; and the other is using the rapid prototyping system based on MEM-350 to process it, which is mainly to restore the physical 3D-face model based on ABS materials.The research results of the paper produce a positive role for building a 3D-face recognition system with more robustness and developing the other related subjects.更多还原