Research On Geometric Object Fitting Based On Three-Dimensional Point Cloud

Geometry fitting algorithm is one of the important algorithms to detect whether product specifications are qualified.Its main task is to calculate product parameters and detect whether product specifications are qualified according to three-dimensional point cloud data,one of the expression forms of product digitalization.Most geometry is composed of regular surfaces,therefore,the geometry fitting process is actually a collection of regular surface fitting processes.Typical surface fitting algorithms mainly include least squares regular surface fitting algorithm,feature-based regular surface fitting algorithm,and learning-based regular surface fitting algorithm.However,the existing algorithms have some shortcomings in fitting accuracy and fitting methods.In view of the shortcomings of the above algorithm in the automatic detection of product quality,an efficient and accurate geometry fitting algorithm is proposed.Firstly,the algorithm uses the idea of deep learning to efficiently classify geometry types according to three-dimensional point cloud of the geometry.The idea can be summarized in the following three steps:(1)it generates three-dimensional point cloud dataset Regular Surface Net containing eight types of geometry;(2)it designes an efficient neural network Fit Net to extract point cloud features from the geometry;(3)it uses the Regular Surface Net dataset to classify geometry types.Secondly,the algorithm uses the idea of generalized total least squares to accurately fit the geometry parameters according to three-dimensional point cloud of the geometry.The idea can be summarized into the following three steps:(1)it lists the equation expressions of a variety of regular surfaces;(2)it divides the geometry into multiple regular surfaces according to the classification results of Fit Net;(3)it uses the generalized total least square method to fit the parameters of multiple surfaces and integrate them into geometric parameters.Extensive experiments validate that the geometry fitting algorithm proposed in this thesis can efficiently and accurately fit the types and parameters of the geometry.