基于三维点云智能配准的船体分段建造精度控制方法研究

基于三维点云的船体分段建造精度控制方法能够极大地降低精度测量的时间消耗与人力消耗，对于提高造船流程的信息化与智能化具有重要意义。然而，传统点云处理算法受限于人工设计的描述方法和简单的最近点对应原则，对于船体分段三维扫描的复杂点云数据，难以提供高效合理的粗配准结果和意义明确的精配准对应。因此，本研究结合船体分段复杂结构特点，基于分层超体素分割和PointNet三维卷积网络的深度学习方法，对测度点云进行高效准确的构件分块识别，设计基于显著构件组合的粗配准算法与基于概率对应的精配准算法，充分考虑工艺约束、测量点云固有信息、点云局部语义信息，返回精确合理的船体分段扫描三维点云-CAD配准结果与误差评估。本项目将有力推动工业点云配准理论的智能化进程，研究成果可广泛应用于船舶制造、城市建筑、大型机械工件等大尺寸工业产品的误差测量中。

The precision control of hull block construction based on 3D point cloud can greatly reduce the time and manpower consumption of accuracy measurement, which is of great significance for improving the informationization and intelligence of shipbuilding process.However, the traditional point cloud processing algorithm is limited by the handcrafted descriptive method and the naive nearest point correspondence principle, and it is difficult to provide efficient and reasonable rough registration results and fine registration results with clear meaning.Therefore, this project considers the complex characteristics of hull blocks and performs efficient and accurate semantic segmentation based on hierarchical super-voxel segmentation and PointNet convolution network. Via efficient segmentation and identification of the measurement point cloud, a rough registration algorithm based on significant component combination and a fine registration algorithm based on probability correspondence are designed.The algorithm fully considers the process constraints, the measurement point cloud inherent information, the point cloud local semantic information, and returns accurate cloud-CAD registration results of hull blocks.This project will promote the intelligent process of industrial point cloud registration theory.The research results can be widely used in the error analysis of large-scale industrial products such as shipbuilding, urban construction and large mechanical parts.