3D reconstruction from uncalibrated images has many applications. This paper proposed a 3D reconstruction method from uncalibrated images. The method was based on factorization and bundle adjustment algorithm. First, it got camera projection matrix and object coordinates in projective space using factorization. Then with orthogonality of rotation matrix and rank 3 of the absolute quadric as the constraint, it upgraded the projective space to euclidean space. Finally, it applied bundle adjustment to refine all the parameters. The inner, exterior and distortion parameters could be computed simultaneously. Simulation experiments show that the mean distance error are about 0. 1530 mm and 0. 6712 mm separately when the reconstruction field is about 1000 mm × 1000 mm ×400 mm and the image detecting error are in 0-1 pixels and 0-2pixels. Real experiment shows that the reconstruction error is below 0. 3 mm at the field of 500 mm ×500 m ×200 mm. The proposed algorithm is robust and can be guide for real engineering application.%由非定标图像重建三维场景有着广泛的应用.给出了一种非定标多视图像三维重建算法.该算法主要基于因子分解和光束法平差技术.首先用因子分解方法得到射影空间下相机投影矩阵和物点坐标,以旋转矩阵的正交性以及对偶绝对二次曲面秩为3为约束,将射影空间升级到欧式空间,最后用光束法平差进行优化.该方法可同时获得相机的内外参数、畸变系数和场景的三维坐标.仿真实验表明,在1000 mm× 1000 mm× 400mm的范围内,当像点检测误差在0-1 pixel和0-2pixel内,所重建三维点的误差分别为0.1530 mm和0.6712 mm.在500 mm×500 m×200 mm下,真实实验重构三维点的误差在0.3 mm以内.所提出的算法稳定可靠,可对实际工程进行指导.
展开▼
