面向RGB-D数据的4D-ICP点云配准方法

4D-ICP point cloud registration method for RGB-D data

针对三维彩色物体的配准问题,提出一种面向RGB-D数据的点云配准方法.首先利用主方向贴合方法将待配准的两片点云快速拉近,使它们近似对齐;在点云精确配准阶段,将RGB颜色值转换成单通道的灰度值,并将灰度值范围映射到几何数据的范围,由映射后的灰度值和点云的几何信息构成四维向量;然后由点的局部邻域几何信息和颜色信息构造混合特征描述子,根据混合特征描述子获得源点云的特征点,在四维向量空间,利用k近邻算法在目标点云中搜索对应点,以提高搜索效率;最后,定义了一种基于4D欧氏距离的ICP算法,通过4D-ICP迭代算法实现点云的精确配准.实验结果表明,面向RGB-D数据的4D-ICP配准方法,能够快速有效地实现RGB-D点云模型的配准,并在配准精度和保持颜色纹理方面效果突出.

Aiming at the registration problem of color objects,apoint cloud registration method for RGB-D data is proposed.Firstly,two point clouds are quickly registered by principal component analysis algorithm to make them roughly aligned.In the fine registration phase,RGB color values are converted into single-channel gray values,and the range of gray values is mapped to the range of geometric data.Four-dimensional vectors are formed from the mapped gray values and the geometric information of the point cloud.Secondly,the mixed features descriptors are constructed by the local neighborhood geometric information and color information.The feature points are obtained according to the mixed features.The k nearest neighbor algorithm is used to search the corresponding points in the4 Dvector space for improving the search efficiency.Finally,a 4 DEuclidean distance-based ICP algorithm is defined,and the fine registration of the point clouds is achieved by 4 D-ICP iterative process.The experimental results show that the 4 D-ICP registration method for RGB-D data can effectively achieve the registration of the RGB-D point cloud model,and has outstanding performance in registration accuracy and color texture preservation.