轨迹空间中基于特征符号搜索算法的非刚体三维运动重建研究

现有研究一般是在正交约束条件下采用最小二乘法来求解三维运动轨迹基重建问题,然而这些算法的性能容易受噪声影响,产生不适定性。为此,通过利用离散线性的方法为三维运动搜索最优轨迹基系数和三维结构,建立了一种基于特征符号搜索算法的三维运动轨迹基重建方法。利用该方法,对一系列单目图像序列进行了重建实验研究,结果表明所提出的方法是可行的,并提高了重建算法的准确性。

Non-rigid 3D Structure Reconstruction in Trajectory Space Based on Feature-sign Search Algorithm

Existing approaches to recover non-rigid structure from motion in trajectory space mainly use the least square method under orthogonality constraint.However,the performance of these algorithms is easily influenced by noise,which makes the system ill-conditioned.Thus,this paper proposes a reconstruction method based on feature-sign search algorithm.The algo- rithm can find the optimal solution of trajectory basis coefficients and 3D structure by using an efficient discrete line search method.With this optimizing strategy,a series of reconstruction experiments are conducted on monocular image sequences. The results show that the algorithm is efficient and accurate.