Reflectional and Rotational Symmetry Detection of CAD Models Based on Point Cloud Processing

Symmetry is the most common attribute present in many of the real world and engineering 3D models. Organizing, managing, and retrieving the 3D Computer-Aided Design (CAD) models for engineering applications is tedious and time-consuming. To simplify the procedure, this work proposes the detection of rotational and reflectional symmetries of CAD models based on point cloud data processing. With the support of view-based shape classification at three orthogonal axes and at regular intervals of rotation, the symmetry of the model is determined. Experiments are performed on the 3D Shape Net benchmark dataset with different labels. Recall, Precision, and F-score are measured for the performance of the methodology applied. Four categories of CAD models are applied for the symmetry detection and a 0.95 F-score is obtained as maximum. As a future scope, the identification and classification of scan models of the reverse-engineered objects will be performed with the application of 3D deep learning of point clouds.