基于曲率的点云自动配准方法

Automatic Registration Method Based on Curvature

由于三维扫描仪采集范围的制约,单次扫描仅得到单一视角的深度图像,需研究深度图像的配准问题实现三维建模。现有商业软件大多需要人工标定实现深度图像配准,为改进此问题,提出一种基于曲率约束的三维点云自动配准法。采用4PCS方法实现两个三维点云模型的自动初配准,用点到面的ICP法及最小二乘法,计算三维模型的刚性变换矩阵。为消除迭代过程中的误匹配,将顶点曲率作为约束,提高了点云配准的准确性。实验结果表明:根据曲率约束去除错误匹配点对后,配准精度提高。

Due to the acquisition range constriction of three-dimensional, a scanner could only obtain a single perspective of the deep image achieving the 3-D modeling from researching the registration of deep image. Currently, most of the existing commercial software requires manual label to achieve the registration of deep image. In order to improve this problem, an automatic registration method based on the constriction of curvature was proposed. At the beginning of registration, the method 4-Points Congruent Sets for Robust Surface Registration(4PCS) was used to achieve the initial and automatic registration. In the phase of accurate registration, ICP and linear least-square optimization method was used to get 3-D model's Rigid transformation matrix. In order to eliminate the iterative process of mismatch problem, the curvature as a constraint was taken to improve the accuracy of point cloud registration. Experiment indicates that after the removal of false match points according to the constraint of curvature, the registration accuracy of the model is increased and perfect.