一种极线近似的双目结构光相位立体匹配方法北大核心CSCD

针对双目结构光三维重建中的相位立体匹配效率较低问题,提出了一种极线近似的快速匹配方法。首先依据两相机光心与左像素形成平面与右成像平面交线对极线进行描述;对每行像素中部分区域的对应极线进行近似,结合立体视觉的连续性约束,使得孤立的沿各自极线搜索匹配相位改为区域内沿近似极线连续搜索匹配相位;同时结合双目相机位姿特性,实现全局均匀分区,避免针对每行数据重复分区;计算过程中采用查表法辅助计算。实验结果表明,立体匹配后获得点云平均误差为0.436 mm,属可接受误差范围内,立体匹配计算速度平均提升10.18倍,对640×480尺寸图像可在17 ms内完成立体匹配,可应用于结构光实时三维重建。

A phase stereo matching method based on integrated epipolar line using binocular structured light

For the low efficiency of phase stereo matching in binocular structured light 3D reconstruction, a fast matching method with polar line approximation is proposed. Firstly, the polar lines are described based on the intersection line between the optical centers of the two cameras and the left pixel forming plane and the right imaging plane; the corresponding polar lines of some regions in each row of pixels are approximated, and the continuity constraint of stereo vision is combined, so that the isolated search for matching phase along the respective polar lines is replaced by the continuous search for matching phase along the approximate polar lines in the region; meanwhile, the global uniform partitioning is realized by combining the bit pose characteristics of binocular cameras to avoid repeated partitioning for each row of pixels; the calculation is assisted by the table look-up method. The global uniform partitioning is combined with the binocular camera pose characteristics to avoid repeated partitioning for each row of data. The experiments show that the average error of the point cloud obtained after stereo matching is 0.436 mm, which is within the acceptable error range, and the average speed of stereo matching calculation is increased by 10.18 times.