基于连续相位稠密匹配的结构光三维重建优化方法

An Optimization Method for Structured-light Three-dimensional Reconstruction Based on Dense Continuous-phase Matching

提出了一种基于连续相位稠密匹配的结构光三维(3D)重建优化方法。利用标定的相机阵列获取投影的结构光变形条纹,利用格雷码划分相位级次,提出等相位约束条件,实现不同相机图像之间的稠密匹配;然后利用空间3D点与相机阵列之间的投影关系,建立各相机位姿与重建点云全局优化模型;利用观测值反向投影的欧式距离建立非线性优化的目标函数,并将此优化问题转为李代数上的图优化模型进行迭代求解。利用公开数据集和实测数据进行实验,结果表明,所提方法可以实现360°无遮挡的3D重建,且重建精度相对传统无优化结构光方式提高了6.94%,相机位姿精度提高了9.77%。所提方法相比参考方法对相机畸变标定精度要求低,并可用于模型点云的非刚体融合。同时由于相机阵列的使用,重建过程中目标物体无需多次旋转,使得物体的3D重建更加方便。

An optimization method for structured-light three-demensional(3 D)reconstruction based on dense continuous-phase matching is proposed.Firstly,the deformed structured-light pattern is captured by calibrated camera array,and the different camera images are matched densely based on the proposed constraint condition of phase determined Gray-code strip.Secondly,a global optimization model of the camera poses and reconstruction point clouds is established under the projection relationship between spatial 3 D points and camera array.Thirdly,the cost function of the nonlinear model is designed based on the observed Euclid distance of back projection,and the iteration result is minimized by graph optimization method with Lie-algebra.Experiments on both public and real datasets show that the proposed optimization method is helpful to obtain the 360°reconstruction without occlusion,and the reconstruction accuracy is improved about 6.94%,and the camera poses accuracy is improved by about 9.77%.The requirement of calibration accuracy for camera distortion is not very strict,and the proposed method can be applied in the non-rigid fusion of point cloud models.Meanwhile,it is more convenient without multiple times rotations.