基于点云识别与参数化建模的残缺物体重建方法

打开文本图片集

中图分类号：TN249⁃34 文献标识码： A 文章编号：1004⁃373X（2025）12⁃0105⁃06

Abstract：Although laser scanning point cloud technology has been widely used in the field of 3D modeling， there is also a common problem of missing parts， which makes it difficult to reconstruct high ⁃ precision solid models. The existing modeling methods are expensive， and when faced with a large number of models of the same type in actual industrial scenarios， the consumption of human resources is excessive. Although the model library replacement scheme can improve efficiency， the generated models often have a mismatch with the original point cloud ratio， making it difficult to meet actual production needs. On this basis， a method of parametric inverse modeling based on object category recognition is proposed， which can effectively solve the problem of missing parts and dynamically adjust the size to meet the needs of different scenarios. In this method， deep learning networks are firstly used to segment and classify incomplete point clouds， obtaining the categories and models of incomplete objects. The bounding box calculation is used to obtain the size and position information of objects. The parameterized driving and assembly for different components of the object are conducted by means of 3D modeling to realize the high⁃precision reverse modeling of the solid model. The similarity detection is conducted by means of iterative closest point （ICP） registration algorithm to ensure the geometric consistency between the reconstructed model and the original point cloud.

Keywords： laser point cloud; parametric modeling; inverse modeling; deep learning; incomplete point cloud; ICP registration

0 引 言

在点云数据建模方面，点云数据由大量的三维坐标点组成，能够精确地反映物体的几何形态，是目前获取物体表面形状信息的重要手段。（剩余7988字）