视差松弛传播约束的倾斜影像特征匹配方法

Oblique Images Feature Matching Method Based on Parallax Relaxation Transmission Constraints

航空倾斜影像已成为大规模城市三维重建的主要数据源,然而由于倾斜影像存在较大的俯仰、横滚角,影像之间差异较大,其自动特征匹配仍存在明显的可靠性问题,交互式转刺同名点仍必不可少。已有交互式同名点方法通常仅依赖于核几何信息,其约束相对较弱,半自动匹配成功率较低。提出一种顾及视差松弛传播约束的倾斜影像核几何特征匹配方法,基于视差连续性假设,通过已知同名点拟合一张可近似表达物方视差信息的薄板样条函数,并结合核线信息,确定合适的搜索区域。对搜索区域按距离可靠性排序,根据几何顺序进行模板匹配,直至相关系数满足阈值。经试验证明,本文方法较传统核几何特征匹配方法,其匹配速度、可靠性均有显著提升,为倾斜影像后续应用提供保障。

Aerial oblique image has become the main data source of large-scale urban 3D reconstruction, but there are large pitch angle, roll angle and big diferences between oblique images. There are still obvious reliability problems in automatic feature matching of oblique images, and the interactive tie points are essential. Existing interactive matching methods usually only depend on epipolar geometrical information, and the constraints are relatively weak, the success rate of semi-automatic matching is low. Therefore,this paper presents a kind of oblique images feature matching method based on parallax relaxation transmission constraints. Based on the continuity hypothesis of parallax,a thin plate spline function which can express approximately the object parallax information is ftted by known tie points, and combined with the epipolar information to determine appropriate search area. The pixels in the search area are sorted by distance reliability, and the template matching is carried out in geometric order until the correlation coefficient satisfies the threshold. Experiments show that the matching speed and reliability of the proposed method are signifcantly improved compared with the traditional feature matching method, which provides a guarantee for subsequent application of oblique images.