一种透视投影下的非刚体迭代重构方法
|
摘要
透视投影下的非刚体三维重构中既存在着非刚体形变也存在着透视形变,同时区分两种形变是比较困难的。大多数非刚体三维重构算法通常进行了简化,只考虑正交投影下的非刚体重构。为了解决透视投影下的非刚体重构问题,将基于形状基的EM算法和仿射迭代方法进行有机结合,提出了一种适用于透视投影的新算法。该算法利用EM算法可解决数据点丢失问题的优势以及仿射迭代方法结果较为精确的优点,可提高计算的鲁棒性和重构效果。为验证算法的有效性,选择5 000帧面部表情连续变化的Frank图像序列进行了迭代重构仿真实验。实验结果表明,所提出算法在透视投影模型下具有较好的鲁棒性以及重构效果,能准确地实现动态场景下对人脸的非刚体三维结构重构。
Non-rigid three-dimensional reconstruction under perspective projection has both non-rigid deformation and perspective deformation,and it is difficult to distinguish them.Most of the non-rigid three-dimensional reconstruction algorithm is usually simplified,only considering the non-rigid body reconstruction in orthogonal projection.In order to solve the problem of non-rigid body reconstruction in perspective projection,after combination of the EM algorithm based on shape basis and the affine iterative method,a new algorithm for perspective projection is proposed.It uses the advantage of EM algorithm that can solve the problem of data point loss and of affine iterative method that is more accurate and thus can improve the robustness and reconstruction effect.In order to its effectiveness,an iterative reconstruction simulation experiment has been performed by selecting 5 000 consecutive Frank images with facial expressions,which indicates that it has good robustness and reconstruction effect under the perspective projection model,and can accurately reconstruct the nonrigid three-dimensional structure of human face under dynamic scene.
Non-rigid three-dimensional reconstruction under perspective projection has both non-rigid deformation and perspective deformation,and it is difficult to distinguish them.Most of the non-rigid three-dimensional reconstruction algorithm is usually simplified,only considering the non-rigid body reconstruction in orthogonal projection.In order to solve the problem of non-rigid body reconstruction in perspective projection,after combination of the EM algorithm based on shape basis and the affine iterative method,a new algorithm for perspective projection is proposed.It uses the advantage of EM algorithm that can solve the problem of data point loss and of affine iterative method that is more accurate and thus can improve the robustness and reconstruction effect.In order to its effectiveness,an iterative reconstruction simulation experiment has been performed by selecting 5 000 consecutive Frank images with facial expressions,which indicates that it has good robustness and reconstruction effect under the perspective projection model,and can accurately reconstruct the nonrigid three-dimensional structure of human face under dynamic scene.
引文
[1]Bregler C,Hertzmann A,Biermann H.Recovering non-rigid3D shape from image streams[C]//IEEE conference on computer vision and pattern recognition.[s.l.]:IEEE,2000:690-696.
[2]Torresani L,Hertzmann A,Bregler C.Learning non-rigid 3D shape from 2D motion[C]//NIPS.[s.l.]:[s.n.],2004:1555-1562.
[3]Gonzalez-Mora J,Torre F D L,Guil N,et al.Learning a generic 3D face model from 2D image databases using incremental structure-from-motion[J].Image and Vision Computing,2010,28(7):1117-1129.
[4]LladóX,Bue A D,Oliver A,et al.Reconstruction of non-rigid3D shapes from stereo-motion[J].Pattern Recognition Letters,2011,32(7):1020-1028.
[5]Khan I.Non-rigid structure-from-motion with uniqueness constraint and low rank matrix fitting factorization[J].IEEE Transactions on Multimedia,2014,16(5):1350-1357.
[6]林义闽,吕乃光,娄小平,等.用于弱纹理场景三维重建的机器人视觉系统[J].光学精密工程,2015,23(2):540-549.
[7]彭亚丽,刘侍刚,裘国永.一种线性迭代非刚体射影重建方法[J].西安交通大学学报,2015,49(1):102-106.
[8]吴悦,翁小兰.非刚体三维运动图像重建优化模型仿真[J].微电子学与计算机,2015,32(5):157-160.
[9]李水平,吴雨.一种融合纹理的三维图像重建快速实现方法[J].计算机技术与发展,2014,24(5):138-141.
[10]闫晓萌.基于稀疏逼近的非刚体三维运动重建研究[D].杭州:浙江理工大学,2016.
[11]张珊珊,吕东辉,孙九爱.近光源光度立体三维重建误差分析[J].计算机技术与发展,2015,25(1):168-172.
[12]彭亚丽,刘侍刚,贲晛烨,等.基于非刚体的线性迭代相机自标定方法[J].电子学报,2016,44(5):1051-1054.
[13]刘彦宏,王洪斌,杜威,等.基于图像的树类物体的三维重建[J].计算机学报,2002,25(9):930-935.
[14]Christy S,Horaud R.Euclidean shape and motion from multiple perspective views by affine iterations[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1996,18(11):1098-1104.
[2]Torresani L,Hertzmann A,Bregler C.Learning non-rigid 3D shape from 2D motion[C]//NIPS.[s.l.]:[s.n.],2004:1555-1562.
[3]Gonzalez-Mora J,Torre F D L,Guil N,et al.Learning a generic 3D face model from 2D image databases using incremental structure-from-motion[J].Image and Vision Computing,2010,28(7):1117-1129.
[4]LladóX,Bue A D,Oliver A,et al.Reconstruction of non-rigid3D shapes from stereo-motion[J].Pattern Recognition Letters,2011,32(7):1020-1028.
[5]Khan I.Non-rigid structure-from-motion with uniqueness constraint and low rank matrix fitting factorization[J].IEEE Transactions on Multimedia,2014,16(5):1350-1357.
[6]林义闽,吕乃光,娄小平,等.用于弱纹理场景三维重建的机器人视觉系统[J].光学精密工程,2015,23(2):540-549.
[7]彭亚丽,刘侍刚,裘国永.一种线性迭代非刚体射影重建方法[J].西安交通大学学报,2015,49(1):102-106.
[8]吴悦,翁小兰.非刚体三维运动图像重建优化模型仿真[J].微电子学与计算机,2015,32(5):157-160.
[9]李水平,吴雨.一种融合纹理的三维图像重建快速实现方法[J].计算机技术与发展,2014,24(5):138-141.
[10]闫晓萌.基于稀疏逼近的非刚体三维运动重建研究[D].杭州:浙江理工大学,2016.
[11]张珊珊,吕东辉,孙九爱.近光源光度立体三维重建误差分析[J].计算机技术与发展,2015,25(1):168-172.
[12]彭亚丽,刘侍刚,贲晛烨,等.基于非刚体的线性迭代相机自标定方法[J].电子学报,2016,44(5):1051-1054.
[13]刘彦宏,王洪斌,杜威,等.基于图像的树类物体的三维重建[J].计算机学报,2002,25(9):930-935.
[14]Christy S,Horaud R.Euclidean shape and motion from multiple perspective views by affine iterations[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1996,18(11):1098-1104.