带线约束的摄影测量网格变分精化算法

针对城市场景三维重建模型在有线特征的边缘区域过于平滑的问题,提出一种带线约束的摄影测量网格变分精化方法。算法以初始重建网格模型为基础,引入3个能量项,将网格精化问题转化成能量下降问题。首先结合所有的影像信息构建影像一致性约束项,然后对网格表面顶点附加正则化约束,最后引入三维线特征约束,将3个能量项加权相加后离散化到每个顶点,得到梯度变化值。采用梯度下降法,使顶点沿着梯度方向移动,当能量不再下降或迭代一定次数时,即得到了精化后的网格模型。试验结果表明,本文算法能较好地保持边缘特征,与现有的泊松重建算法相比,网格的质量更高,视觉效果更好。

Variational refinement of mesh with line constraint for photogrammetry

In order to solve the problem of that urban 3D reconstruction is too smooth in the edge area with line features, it is proposed a variational refinement of mesh with line constraint. The algorithm takes the initial reconstruction mesh as input and introduces three energy terms. Transforming the refinement problem into energy reduction problem. Firstly, photo-consistency constraints are constructed by combining all image information, and then regularization constraints are added to all the vertices of the mesh. Finally, 3D line constraint is added to energy term. The gradient difference value is obtained by discretizing sum of three weighted energy term to each vertex. Using gradient descent method to make each vertex move along the gradient vector. When the energy no longer decreases or reaches enough iterations, the refined mesh model is obtained. Calculating the coordinate corrections of each vertex by variational refinement iteration method. Thus, vertices naturally migrate to the object edges if any. The results show that the proposed algorithm can preserve the edge features better. Compared with the existing Poisson surface reconstruction algorithm, the quality of mesh is higher and the visual effect is better.