Object Geometric Primitives Detection from 3D Point Clouds Based on the Profiles on Cutting Planes
Liyong Liu, Xiaowei Tu, Qinghua Yang, Jingyuan Yu · 2021
Common man made objects can be represented or composed by some geometric primitives, such as planes, cylinders, spheres, cones, etc. Existing methods reconstruct the geometric primitives by fitting the object with whole point clouds, and these methods need to set the types of model in advance. In this paper, a geometric primitives modeling method based on cutting-plane is proposed for dense 3D point clouds without setting model type in advance. Our method assumes that the original object point clouds have been segmented. First, we cut repeatedly point clouds along the z-axis with a cutting plane, which is parallel to the XOY plane with a step. In this paper, the line of sight is set as Z-axis, because sufficient contours information can be obtained in such cutting direction. The contours may contain both straight lines and quadric curves, so we use clustering to separate lines and curves. After repetitive cutting, each curve cluster has many same-type curves. Then we can recognize the type and estimate the parameters of the point clouds by calculating the curves changing relationship of curves clustering. Last we evaluated our algorithm using synthetic data and real data, and the result shows that our proposed algorithm is efficient and robust.