基于多Kinect的三维人脸重建研究
|
摘要
人脸三维数字化在医疗、影视制作和虚拟现实等领域具有重要意义,是计算机视觉研究的热点.为快速、完整、低成本和精确地实现人脸三维数字化,提出一种基于多Kinect的三维数字化系统.首先,连续采集8帧点云信息,避免信息缺失,改进双边滤波算法,对Kinect获取的点云滤波去噪;然后,精简点云拼接区域,先采用点特征直方图在两点云间寻找配准点,构造协方差矩阵,奇异值分解法求解该矩阵,得到初始变换参数,粗拼接点云;再通过改进的共享最近邻居聚类算法加快粗拼接后两点云间的最近点搜索速度,利用改进的最近点迭代算法实现精拼接;最后,实验结果表明:该系统能够快速、完整和精确地三维数字化人脸,达到了实验预期效果.
Three-dimensional face digitization in medical,video making,virtual reality and other fields has great significance.It is the focus of computer vision research.In order to quickly realize three-dimensional digital face with low cost and high precision,a 3 D digitization experimental system with three Kinect devices is built.Firstly,continuously collect 8 frames of point cloud information to solve the problem of missing data improve the bilateral filtering algorithm,and remove noise about Kinect pick-up information.Then,the point cloud stitching area is simplified.After this step,the point feature histogram(PFH)is used to find the registration point between the two point clouds,and construct their covariance matrix.Singular value decomposition method is used to solve the matrix,and obtain the initial transformation parameters,so the point cloud is roughly matched.Subsequently,the refined shared nearest neighbor clustering algorithm(RSNN)is used to speed up the nearest point searching between two point clouds,and the improved nearest point iterative algorithm is used to realize the accurate registration.Finally,the experimental results show that the system can quickly and accurately digitize complete face with low-cost.It achieves the expected effect.
Three-dimensional face digitization in medical,video making,virtual reality and other fields has great significance.It is the focus of computer vision research.In order to quickly realize three-dimensional digital face with low cost and high precision,a 3 D digitization experimental system with three Kinect devices is built.Firstly,continuously collect 8 frames of point cloud information to solve the problem of missing data improve the bilateral filtering algorithm,and remove noise about Kinect pick-up information.Then,the point cloud stitching area is simplified.After this step,the point feature histogram(PFH)is used to find the registration point between the two point clouds,and construct their covariance matrix.Singular value decomposition method is used to solve the matrix,and obtain the initial transformation parameters,so the point cloud is roughly matched.Subsequently,the refined shared nearest neighbor clustering algorithm(RSNN)is used to speed up the nearest point searching between two point clouds,and the improved nearest point iterative algorithm is used to realize the accurate registration.Finally,the experimental results show that the system can quickly and accurately digitize complete face with low-cost.It achieves the expected effect.
引文
[1]于爱华,白煌,孙斌斌,等.基于优化投影矩阵的人脸识别技术研究[J].浙江工业大学学报,2016,44(4):392-398.
[2]陈志杨,罗飞.基于WebGL的三维建筑模型重建[J].浙江工业大学学报,2016,44(6):608-613.
[3]范哲.基于Kinect的三维重建[D].西安:西安电子科技大学,2012.
[4]WEBER T,HANSCH R,HELLWICH O.Automatic registration of unordered point clouds acquired by Kinect sensors using an overlap heuristic[J].ISPRS journal of photogrammetry and remote sensing,2015,10(2):96-109.
[5]CHEN G,LI J T,WANG B,et al.Reconstruction 3Dhuman models with a Kinect[J].Computer animation and virtual worlds,2016,27(1):72-85.
[6]TONG J,ZHOU J,LIU L G,et al.Scanning 3Dfull human bodies using Kinects[J].IEEE transactions on visualization and computer graphics,2012,18(4):643-650.
[7]JIA W D,YI W J,SANIIE J,et al.3Dimage reconstruction and human body tracking using stereo vision and Kinect technology[C]//2012IEEE International Conference on Electro/Information Technology.Indianapolis,USA:University Place Conference Center,2012:1-4.
[8]CHEN C M,ZOU W,WANG J J.3Dsurface reconstruction based on Kinect[C]//2013Third International Conference on Information Science and Technology.Yangzhou,China:IEEE,2013:986-990.
[9]LI L,MIAO Z J,LIANG M G.3Dreconstruction based on Kinect[C]//2014 12th IEEE International Conference on Signal Processing.Hangzhou,China:IEEE,2014:1247-1250.
[10]李方召,金士尧,林帅,等.1秒采集的三维人脸头像摄影系统[J].计算机辅助设计与图形学学报,2015,27(6):1099-1103.
[11]董国胜.基于Kinect的三维彩色人体表面重建技术的研究[D].上海:东华大学,2015.
[12]管秋,金俊杰,张剑华,等.基于最优RANSAC算法的非增加式多视图三维重建[J].浙江工业大学学报,2015,43(5):473-478.
[13]陶海跻.三维点云配准算法研究[D].南京:东南大学,2014.
[14]王小超.三维数字几何处理中特征提取与孔洞修补研究[D].大连:大连理工大学,2014.
[15]杨帆,白宝兴,张振普,等.基于公共标志点的三维点云拼接方法研究[J].长春理工大学学报(自然科学版),2014,37(2):76-79.
[16]伍梦琦.无标记点三维点云数据拼接技术的研究[D].武汉:华中科技大学,2015.
[17]BESL P J,MCKAY N D.A method for registration of 3-D shapes[J].IEEE transactions on pattern analysis and machine intelligence,1992,14(2):239-256.
[18]左超,鲁敏,谭志国,等.一种新的点云拼接算法[J].中国激光,2012,39(12):211-218.
[19]LI S H,WANG J X,LIANG Z Q,et al.Tree point clouds registration using an improved ICP algorithm based on KDTree[J].IEEE international geoscience and remote sensing symposium,2016,11(9):4545-4548.
[20]RUSU R B,MARTON Z C,BLODOW N,et al.Learning informative point classes for the acquisition of object model maps[C]//200810th IEEE International Conference on Control,Automation,Robotics and Vision.Hanoi,Viemtnam:IEEE,2008:588-1177.
[21]KUMARI S,MAURYA S,GOYAL P,et al.Scalable parallel algorithms for shared nearest neighbor clustering[C]//2016IEEE 23rd International Conference on High Performance Computing.Hyderabad,India:IEEE,2016:72-81.
[22]吴诗婳,吴一全,周建江,等.基于2维灰度熵阈值选取快速迭代的图像分割[J].中国图象图形学报,2015,20(8):1042-1050.
[2]陈志杨,罗飞.基于WebGL的三维建筑模型重建[J].浙江工业大学学报,2016,44(6):608-613.
[3]范哲.基于Kinect的三维重建[D].西安:西安电子科技大学,2012.
[4]WEBER T,HANSCH R,HELLWICH O.Automatic registration of unordered point clouds acquired by Kinect sensors using an overlap heuristic[J].ISPRS journal of photogrammetry and remote sensing,2015,10(2):96-109.
[5]CHEN G,LI J T,WANG B,et al.Reconstruction 3Dhuman models with a Kinect[J].Computer animation and virtual worlds,2016,27(1):72-85.
[6]TONG J,ZHOU J,LIU L G,et al.Scanning 3Dfull human bodies using Kinects[J].IEEE transactions on visualization and computer graphics,2012,18(4):643-650.
[7]JIA W D,YI W J,SANIIE J,et al.3Dimage reconstruction and human body tracking using stereo vision and Kinect technology[C]//2012IEEE International Conference on Electro/Information Technology.Indianapolis,USA:University Place Conference Center,2012:1-4.
[8]CHEN C M,ZOU W,WANG J J.3Dsurface reconstruction based on Kinect[C]//2013Third International Conference on Information Science and Technology.Yangzhou,China:IEEE,2013:986-990.
[9]LI L,MIAO Z J,LIANG M G.3Dreconstruction based on Kinect[C]//2014 12th IEEE International Conference on Signal Processing.Hangzhou,China:IEEE,2014:1247-1250.
[10]李方召,金士尧,林帅,等.1秒采集的三维人脸头像摄影系统[J].计算机辅助设计与图形学学报,2015,27(6):1099-1103.
[11]董国胜.基于Kinect的三维彩色人体表面重建技术的研究[D].上海:东华大学,2015.
[12]管秋,金俊杰,张剑华,等.基于最优RANSAC算法的非增加式多视图三维重建[J].浙江工业大学学报,2015,43(5):473-478.
[13]陶海跻.三维点云配准算法研究[D].南京:东南大学,2014.
[14]王小超.三维数字几何处理中特征提取与孔洞修补研究[D].大连:大连理工大学,2014.
[15]杨帆,白宝兴,张振普,等.基于公共标志点的三维点云拼接方法研究[J].长春理工大学学报(自然科学版),2014,37(2):76-79.
[16]伍梦琦.无标记点三维点云数据拼接技术的研究[D].武汉:华中科技大学,2015.
[17]BESL P J,MCKAY N D.A method for registration of 3-D shapes[J].IEEE transactions on pattern analysis and machine intelligence,1992,14(2):239-256.
[18]左超,鲁敏,谭志国,等.一种新的点云拼接算法[J].中国激光,2012,39(12):211-218.
[19]LI S H,WANG J X,LIANG Z Q,et al.Tree point clouds registration using an improved ICP algorithm based on KDTree[J].IEEE international geoscience and remote sensing symposium,2016,11(9):4545-4548.
[20]RUSU R B,MARTON Z C,BLODOW N,et al.Learning informative point classes for the acquisition of object model maps[C]//200810th IEEE International Conference on Control,Automation,Robotics and Vision.Hanoi,Viemtnam:IEEE,2008:588-1177.
[21]KUMARI S,MAURYA S,GOYAL P,et al.Scalable parallel algorithms for shared nearest neighbor clustering[C]//2016IEEE 23rd International Conference on High Performance Computing.Hyderabad,India:IEEE,2016:72-81.
[22]吴诗婳,吴一全,周建江,等.基于2维灰度熵阈值选取快速迭代的图像分割[J].中国图象图形学报,2015,20(8):1042-1050.