Study On Learning-based Methods For Extracting Feature Lines From Meshes

With the development of science and technology,reverse engineering has been widely used in the design and manufacture of mechanical products.The 3d digital geometric mesh model obtained from laser point cloud needs the geometric feature processing such as entity reconstruction,feature recognition and feature editing to be used effectively.The characteristics of the complex grid model intelligent recognition is an important development direction in the field of reverse engineering,the characteristic identification is the basis of feature recognition and extraction of the border,is the basis of the subsequent application of complex triangular mesh model,so how to efficiently extract the characteristics of triangular mesh model line has important research significance.Traditional feature edge extraction USES a single threshold value and decision rules,which are difficult to identify and take out the feature edge that meets the actual requirements.Based on the deep analysis of the geometric features of the feature boundary line and the method of machine learning,a method of feature edge extraction based on learning triangular mesh model is proposed and implemented.The method transforms the feature boundary extraction into the classification problem of triangular edge.In this paper,a 17 dimensional eigenvector composed of two sides of triangle edge,neighborhood curvature of edge vertex and shape diameter is analyzed and constructed.Through manual labeling,the feature vector training data set is obtained,and the general bp-adaboost classifier is trained to obtain the classifier that can recognize the feature boundary.Feature edge recognition is performed on the identified triangular mesh model.The result of identification is proved to be satisfactory by an example.Second,the initial characteristic line,the identification results are generated in the characteristic curve of the initial shear and smooth,and then use the grid edge weights connecting method of characteristic line closure,and select the main contour model connection method of characteristic line is optimized by characteristic line.Finally,the obtained characteristic line and the software obtained characteristic line are compared and analyzed.The feasibility and validity of the method proposed in this paper are verified by an example test.In this paper,the geometric feature calculation,training sample construction and recognition result processing of triangular mesh model are realized by programming on Visual C++ software development platform and Matlab platform.This work laid the foundation for model reconstruction and design feature editing.