基于区域生长的稠密三维点云重建方法

Region-growing approach to reconstruct dense 3Dpoint clouds

目前存在的稀疏方法重建出的三维点云不够充足,导致无法重建出有效三维模型,为此提出一种基于区域生长的方法实现稠密三维点云的重建。以图像特征点匹配作为初始种子,以零均值归一化互相关系数和双视几何约束作为生长过程中图像对应点的匹配标准,在整幅图像上进行匹配生长,采取视差梯度约束和置信度约束进行匹配优化,获得稠密的图像对应点匹配;采用一种基于平面的相机标定方法计算相机的内部参数;利用稠密匹配和相机的内部参数计算图像间的相机运动参数,以投票策略确定相机运动参数;建立双视几何模型重建点云。实验结果表明,该方法能重建出准确的稠密三维点云。

To solve the problem that the efficient 3Dmodel can not be reconstructed because of the insufficiency of reconstructed3 Dpoint clouds using the sparse method,a region-growing approach for the dense 3Dpoint clouds reconstruction was proposed.The growing procedure was started from the seeds derived from the image feature matches and grown with the match criterion called a zero-mean normalized cross-correlation method and involved two match optimization methods including disparity gradient limit and confidence measure limit through the whole image,ended with the dense correspondence matches.A planar-based camera calibration was conducted to obtain the camera＇s interior parameters.The robust camera motion parameters were computed by the dense correspond matches with a vote mechanism.The two-view geometry model was established to reconstruct the 3D point clouds.Accurate and dense 3Dpoint clouds were achieved using the region-growing method.