无序图像三维重构中的图像选择技术

针对无序图像的三维重构问题,本文提出了一种图像选择算法,在三维重构算法运行之前,首先利用图像间的全局运动估计,去除冗余的图像,有效的减少了问题的规模,同时保持图像间的大部分连接关系,并能够较好地防止出现退化情况。为了验证本文的算法,针对一系列图像进行了测试,实验结果证明,作为三维重构的一个预处理过程,提高了三维重构算法的运行效率和算法的鲁棒性,具有很高的实用价值。

Image selection for 3D reconstruction from unordered views

An image selection algorithm is proposed that makes 3D reconstruction from unordered views more effectively. Based on the global motion estimation between image pairs, we remove most of the redundancy images, which keeps the size of problem manageable and avoids loss of connectivity. Further more; ill-posed problems caused by insufficient motion can be avoided after image selection. To verify the effectiveness of the proposed algorithm, we tested it on number of image collections. Experimental result shows as a preprocessor of 3D reconstruction, it can reduce the elapsed time and improves the robustness of the final reconstruction.