光电子技术, 2016, 36 (4): 237, 网络出版: 2016-12-21
基于双目视觉的三维重建和拼接技术研究
Research on Three-dimensional Reconstruction and Stitching Based on Binocular Stereo Vision
机器视觉 三维重建 点云拼接 SURF特征 machine vision three-dimensional reconstruction point cloud stitching SURF feature
摘要
针对自由移动双目视觉的三维重建问题, 提出了一种稠密匹配和稀疏特征匹配相结合的三维重建和点云拼接技术。首先, 在对极几何约束的条件下, 采用绝对值差和作为匹配度量函数对双目视觉左右图像上的图像点进行逐一匹配, 并结合双目视觉内外参数获得各个位置处的稠密三维点云坐标。其次, 对相邻两次移动位置处采集的双目视觉图像进行SURF (Speed-up robust features)特征提取和匹配, 分别计算匹配SURF特征点在不同位置下的三维坐标, 进而求解两个采集位置之间的旋转矩阵和平移矢量, 以实现相邻两位置三维点云的坐标统一。最后, 在室内采用双目视觉实现了两个视点数据的三维重建和拼接。实验结果表明, 文中算法具有计算简单、点云拼接精度较高的优点, 可以应用于大场景三维数据的重建工作。
Abstract
Aiming at reconstruction of freely-moved binocular stereo vision, a novel method was proposed, in which dense matching was used for 3D reconstruction while sparse feature matching for point cloud stitching. Firstly, under conditions of epipolar geometry constraints, the sum of the absolute value of difference was used as a cost function to match the image points in left image and right one, and combined with the internal and external parameters of binocular vision, the dense 3D point cloud coordinates were obtained respectively in different positions. Secondly, the SURF feature points were extracted and matched on left and right image and the corresponding 3D coordinates were obtained at two adjacent moving positions. Then the rotation matrix and translation vector were computed, therefore two adjacent 3D point cloud stitching could be completed. Finally, the reconstruction and stitching of the 3D point cloud at two viewpoints indoors were realized through stereo vision. Experimental results show that this algorithm could be applied to the reconstruction of 3D data in large scenes because of simple calculation and reliable stitching precision.
吕耀文, 康凯. 基于双目视觉的三维重建和拼接技术研究[J]. 光电子技术, 2016, 36(4): 237. LV Yaowen, KANG Kai. Research on Three-dimensional Reconstruction and Stitching Based on Binocular Stereo Vision[J]. Optoelectronic Technology, 2016, 36(4): 237.