基于四目系统的复杂表面高精度密集空间点云获取方法

在敦煌莫高窟文化遗产数字化保护中,针对基于立体视觉的方法对目标进行三维重建时,复杂表面缺乏特征点,难以获取高密度和高精度空间三维点云的问题,提出了一种基于投影和四目立体视觉的方法。首先,采用投影仪向目标投射黑白棋盘格,并检测角点;然后通过软件细分,将棋盘格在横向和纵向分别平移9次,将棋盘格角点数扩展81倍;最后,用基于四目立体视觉系统的匹配策略,短基距系统用来立体匹配,长基距系统用来计算空间坐标,获取均匀、高密度和高精度的三维点云。实验结果表明:与传统的Harris算子相比较,所提方法获取的三维点云数据能更准确地描述目标的三维结构,点数在数量级上有提高。精度方面,所获取三维点云空间坐标的绝对误差小于1 cm,相对误差小于10%。

High precision and Density spatial Point Cloud Acquisition Method for Complex Surface Based on Four Vision System

According to lack of feature points on the complex surface and difficult to obtain high density and high precision spatial 3D point cloud problem when reconstructed the object based on stereo vision in Dunhuang Mogao Grottoes cultural heritage digital protection, a method based on projection and four camera stereo vision system is proposed. Firstly, the projector is used to project the black and white chess board to the object, and the corner points are detected. Then through the software subdivision, the chess board translates nine times in the horizontal and vertical respectively, and the number of chess board corner points extended eighty-one times. Finally, based on the matching strategy of the four camera stereo vision system, the short base distance system is used for stereo matching, and the long base distance system is used to calculate the spatial coordinates, the 3D point cloud with uniform, high density and high precision are obtained. The experimental results show that compared with the traditional Harris operator, the 3D point cloud data obtained by the proposed method can more accurately describe the 3D structure for the object, and the number of points has been improved by magnitude. In precision, the absolute error is less than 1cm, and the relative error is less than 10%.