基于无人机影像的非增量式多视图几何三维重建研究

Non-incremental Multiview Geometry 3D Reconstruction Based on UAV Image

增量式多视图几何三维重建对最初始的视图重建要求高,在后续重建过程中,需要不断的进行捆集调整,因此,最后重建的精度和效率都不高。非增量式三维重建对初始视图重建的精度要求不高,首先求出所有的摄像机参数,通过全局方式一次性计算目标三维点,在重建的精度和效率上要明显优于增量式三维重建。非增量式三维重建的主要难点在于如何精确的求出视图的全局旋转矩阵,为了提高全局旋转矩阵的计算精度,使用最优随机抽样一致方法(A Contrario-Random Sample Consensus,AC-RANSAC)来去除三维重建中错误的二视图关系,与传统的RANSAC设定固定阈值相比,AC-RANSAC能够自适应的设定阈值,避免了不同阈值对重建结果的误差影响。本文编程实现了非增量式多视图几何三维重建,并分别使用增量式和非增量式重建方式对一组无人机影像进行实验。实验结果表明,非增量式三维重建效率高、精度高,适用性更强。

The incremental Multiple view geometry of 3D reconstruction is very dependent on the initial view of the reconstruction and need to constantly adjust bundle.So the final reconstruction precision and efficiency are not precise.Non-incremental 3D reconstruction is not dependent on the initial view of the reconstruction.It finds out all the camera parameters,one-time calculation by means of a global target 3D point,the reconstruction accuracy and efficiency are significantly better than the incremental 3D reconstruction.The mian difficulty of non-incremental 3D reconstruction is how to accurately find out the view of global rotation matrix.In order to improve the calculation accuracy of global rotation matrix,AC-RANSAC algorithm is used to eliminate errors in the reconstruction of the second view.Compared with the traditional RANSAC algorithm to set a fixed threshold,AC-RANSAC can set an adaptive threshold,avoiding the different threshold value with the result of reconstruction error influence.This paper realises the Non-incremental multi-view geometry 3D reconstruction.It respectively uses incremental and non-incremental reconstruction methods to carryout experiments on a group of unmanned aerial vehicle imaging.The experimental results show that the non-incremental 3D reconstruction is more efficient,more accurate and more applicable.