一种线性迭代非刚体射影重建方法

A Linearly Iterative Method for Non-Rigid Projective Reconstruction

为了从非标定图像序列中重建出三维非刚体的射影结构,提出了一种线性迭代非刚体射影重建方法。该方法将所有图像点放入一个图像矩阵中,利用图像矩阵具有低秩的特性,通过因式分解构造一个线性迭代算法来求解图像点的深度因子,最终通过奇异值分解实现非刚体的射影重建。该方法的优点是:射影重建过程都是线性求解,且将所有图像及图像点都平等地对待,并没有倚重某些图像及图像点。模拟实验结果表明,所提重建方法比经典的Alessio方法重投影误差小18%,比Brand方法小27%;最后的真实实验结果表明,所提的重建方法重投影误差只有1.28个像素,从而验证了所提方法的有效性。

A linearly iterative method for reconstruction of 3D non-rigid projection is presented to reconstruct the 3D non-rigid projective structure from an un-calibrated image sequence. All the image points are placed into an image matrix, and the linearly iterative method that employs the singular value decomposition is used to get the depth factors of the image points by using the low rank property of the image matrix. Then the projective reconstruction is realized through the singular value decomposition. The innovations of the proposed method are that the projective reconstruction is linear, all the image points are treated fairly, and it does not rely on certain image or image points. Experiment results with simulation data show that the proposed method is 18% and 27% smaller in re-projective error than Alessio＇s method and Brand＇s method, respectively. A real experiment shows that the method has only 1.28 pixels error, and hence efficient.