以多幅图像非几何约束线段匹配重建建筑物外立面三维线段模型

3DLine Segments Reconstruction for Building Facades with Line Matching Across Multi-Image with Non-Geometry Constraint

针对建筑物外立面三维线段模型重建的现有算法中存在多幅图像线段匹配需要提前获知对极几何约束,以及三维线段重建的过程会因为输入误差的累积而受到影响等问题,提出了一种以非几何约束重建三维线段模型的算法。该算法在多幅图像的线段匹配过程中以点线相结合的高稳定性双视图线段匹配算法为基础,通过匹配线段间的传递性形成匹配线段组,再剔除不相容匹配并合并相容匹配,最终得到准确的匹配线段结果。三维线段重建采用分段重建的方式得到每一个匹配线段组对应的三维线段,再经过对线段长度和角度的约束,剔除不构成建筑物外立面框架的空间线段,最终构成建筑物外立面的三维线段结构。实验结果表明,该算法可以有效重建出建筑物外立面的整体框架。

The previous works on 3D line segment reconstruction of building facades require known geometrical constraints in multi-image line matching, and the line segment reconstruction is affected by accumulated input error. An improved algorithm with non-geometry constraint is proposed. In line matching across multi-image, robust two-view line matching method is chosen, line-point invariants are taken as the foundation and matched-line-segment sets are formed via transitivity of matched line segments. Incompatible matches are eliminated and compatible matches are merged from those sets to form an accurate result. For this line segments reconstruction, 3D line segments corresponding to each matched-line-segment set are obtained in piecewise, and the length constraint and angle constraint are considered to remove the line segments which do not constitute building facades. The whole structure of building facades is thus effectively reconstructed with this algorithm.