基于分段式序列图片集的运动恢复结构
|
摘要
多视图运动恢复结构(Structure from Motion,SFM)是三维重建中相机姿态估计的一种最常用的方法。传统SFM采用增量方式处理图片,算法的时间复杂度是O(n4),当图片数量较多时,重建时间很长。此外,由于图片噪声影响,漂移误差将随着图片数量增加不断累加,影响最终的重建质量。添加集束调整(Bundle Adjustment,BA)可以优化重建结果,但是需要花费更长的时间。在现有增量式算法的基础上,提出基于分段式序列图片集的方法,将序列图片集按照相似度划分为小集合,对每个小集合进行并行计算,减少误差累积量和重建时间,最后再用BA进行全局优化。实验结果表明,该方法能在保持一定精度的前提下,有效减少重建时间。
Structure from Motion(SFM)is a commonly used method of camera pose estimation in 3 D reconstruction.The traditional SFM uses an incremental method to process images, its time complexity is O(n4), and the reconstruction time increases with the number of images. Moreover, due to image noise, drifting errors will accumulate, which directly affects the final reconstruction quality. The reconstruction results can be optimized by adding Bundle Adjustment(BA),but it will cost more time. This method is based on the existing incremental algorithms:divide sequence images into small sets according to the similarity, conduct parallel computing, which will reduce the error accumulation and reconstruction time, and then use the Bundle Adjustment for global optimization. The experimental results show that the method can effectively reduce the reconstruction time while keeping the accuracy.
Structure from Motion(SFM)is a commonly used method of camera pose estimation in 3 D reconstruction.The traditional SFM uses an incremental method to process images, its time complexity is O(n4), and the reconstruction time increases with the number of images. Moreover, due to image noise, drifting errors will accumulate, which directly affects the final reconstruction quality. The reconstruction results can be optimized by adding Bundle Adjustment(BA),but it will cost more time. This method is based on the existing incremental algorithms:divide sequence images into small sets according to the similarity, conduct parallel computing, which will reduce the error accumulation and reconstruction time, and then use the Bundle Adjustment for global optimization. The experimental results show that the method can effectively reduce the reconstruction time while keeping the accuracy.
引文
[1] Noah S,Steven S M,Richard S.Modeling the world from Internet photo collections[J].International Journal of Computer Vision,2008,80(2):189-210.
[2] Noah S,Steven S M,Richard S.Photo tourism:exploring photo collections in 3D[J].ACM Transactions on Graphics,2006,25(3):835-846.
[3] Wu C C.Towards linear-time incremental structure from motion[C]//IEEE International Conference on 3D Vision,2013:127-134.
[4] Schonberger J L,Berg A C,Frahm J M.PAIGE:pairwise image geometry encoding for improved efficiency in structure-from-motion[C]//IEEE Conference on Computer Vision and Pattern Recognition,2015:1009-1018.
[5] Ni K,Dellaert F.HyperSfM[C]//The 2nd International Conference on 3D Imaging,Modeling,Processing,Visualization and Transmission,2012:144-151.
[6] Li H,Hartley R.Five-point motion estimation made easy[C]//IEEE International Conference on Pattern Recognition,2006:630-633.
[7] Quan L.Invariants of six points and projective reconstruction from three uncalibrated images[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,1995,17(1):34-46.
[8] Sinha S N,Steedly D,Szeliski R.A multi-stage linear approach to structure from motion[C]//European Conference on Trends and Topics in Computer Vision.Berlin:Springer-Verlag,2010:267-281.
[9] Cui Z P,Tan P.Global structure-from-motion by similarity averaging[C]//IEEE International Conference on Computer Vision,2015:864-872.
[10] Jiang N,Cui Z P,Tan P.A global linear method for camera pose registration[C]//IEEE International Conference on Computer Vision,2013:481-488.
[11] Moulon P.Global fusion of relative motions for robust,accurate and scalable structure from motion[C]//IEEE International Conference on Computer Vision,2014:3248-3255.
[12] Wilson K,Snabely N.Robust global translations with1DSfM[C]//European Conference on Computer Vision.Springer International Publishing,2014:61-75.
[13] Lowe D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[14] Wu C C,Agarwal S,Curless B,et al.Multicore bundleadjustment[C]//IEEE Conference on Computer Vision and Pattern Recognition,2011:3057-3064.
[15] Chinese Academy of Sciences.(2011).http://vision.ia.ac.cn/zh/data/index.html.
[16] Strecha C,Hansen W V,Gool L V,et al.On benchmarking camera calibration and multi-view stereo for high resolution imagery[C]//IEEE Conference on Computer Vision and Pattern Recognition,2008:1-8.
[17] Furukawa Y,Curless B,Seitz S M,et al.Towards Internetscale multi-view stereo[C]//IEEE Conference on Computer Vision and Pattern Recognition,2010:1434-1441.
[18] Furukawa Y,Ponce J.Accurate,dense,and robust multiview stereopsis[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2007,32(8):1362-1376.
[2] Noah S,Steven S M,Richard S.Photo tourism:exploring photo collections in 3D[J].ACM Transactions on Graphics,2006,25(3):835-846.
[3] Wu C C.Towards linear-time incremental structure from motion[C]//IEEE International Conference on 3D Vision,2013:127-134.
[4] Schonberger J L,Berg A C,Frahm J M.PAIGE:pairwise image geometry encoding for improved efficiency in structure-from-motion[C]//IEEE Conference on Computer Vision and Pattern Recognition,2015:1009-1018.
[5] Ni K,Dellaert F.HyperSfM[C]//The 2nd International Conference on 3D Imaging,Modeling,Processing,Visualization and Transmission,2012:144-151.
[6] Li H,Hartley R.Five-point motion estimation made easy[C]//IEEE International Conference on Pattern Recognition,2006:630-633.
[7] Quan L.Invariants of six points and projective reconstruction from three uncalibrated images[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,1995,17(1):34-46.
[8] Sinha S N,Steedly D,Szeliski R.A multi-stage linear approach to structure from motion[C]//European Conference on Trends and Topics in Computer Vision.Berlin:Springer-Verlag,2010:267-281.
[9] Cui Z P,Tan P.Global structure-from-motion by similarity averaging[C]//IEEE International Conference on Computer Vision,2015:864-872.
[10] Jiang N,Cui Z P,Tan P.A global linear method for camera pose registration[C]//IEEE International Conference on Computer Vision,2013:481-488.
[11] Moulon P.Global fusion of relative motions for robust,accurate and scalable structure from motion[C]//IEEE International Conference on Computer Vision,2014:3248-3255.
[12] Wilson K,Snabely N.Robust global translations with1DSfM[C]//European Conference on Computer Vision.Springer International Publishing,2014:61-75.
[13] Lowe D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[14] Wu C C,Agarwal S,Curless B,et al.Multicore bundleadjustment[C]//IEEE Conference on Computer Vision and Pattern Recognition,2011:3057-3064.
[15] Chinese Academy of Sciences.(2011).http://vision.ia.ac.cn/zh/data/index.html.
[16] Strecha C,Hansen W V,Gool L V,et al.On benchmarking camera calibration and multi-view stereo for high resolution imagery[C]//IEEE Conference on Computer Vision and Pattern Recognition,2008:1-8.
[17] Furukawa Y,Curless B,Seitz S M,et al.Towards Internetscale multi-view stereo[C]//IEEE Conference on Computer Vision and Pattern Recognition,2010:1434-1441.
[18] Furukawa Y,Ponce J.Accurate,dense,and robust multiview stereopsis[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2007,32(8):1362-1376.