3D Geometric Objects Reconstruction Of 2D Geometric Images Based On CNN Classification

Geometric object 3D reconstruction refers to reconstructing a 2D geometric image into a corresponding 3D object,or it also means to restore geometric images to an object in 3D space.3D geometric reconstruction,which has important theoretical research and practical application value in the fields of virtual reality,computer-aided geometric design,education and architecture,and medical image processing,is a branch discipline of 3D reconstruction.Several 3D geometric reconstruction methods are widely used:the regularity-based methods,the deduction-based methods and the divide-and-conquer-based methods.Compared with the other two methods,the regularity-based methods has a more complete theoretical system and faster reconstruction speed,it has been widely concerned by scholars and researchers.In recent years,with the rapid development of convolutional neural network(CNN)technology and its good performance in various fields of computer vision,it has been widely used.The introduction of 3D geometric reconstruction by convolutional neural network(CNN)technology can significantly improve the quality of 3D geometric reconstruction.This paper will focuses on researching 2D geometric images 3D reconstruction in education and architecture field,in order to improve the accuracy of 3D geometric reconstruction and reconstruction speed.After researching previous 3D geometric reconstruction methods,3D geometric objects reconstruction of 2D geometric images based on CNN classification was proposed.This method effectively improves the accuracy and speed of 3D geometric reconstruction.The main work of this paper is reflected in the following aspects: 1.Learn common 3D geometric reconstruction methods,including the regularity-based methods,the deduction-based methods and the divide-and-conquer-based methods.Several representative methods are implemented and compared.2.Focus on the research of CNN technology,and propose a new 3D reconstruction method based on the matching of 2D geometric images and 3D geometric models with CNN classification.First,this method extracts corresponding line drawings of simple or complex 2D geometric images.Secondly,the convolutional neural network technology is introduced to classify the line drawings to realize the high-precision and high-speed recognition of the line drawings.Finally,based on the matching relationship between the line drawings and the 3d geometric models,the 3d geometric model reconstruction of the line drawings is realized.The experimental results show that the method in this paper can effectively improve the recognition accuracy and speed of line drawings,and then reconstruct the corresponding 3D geometric models with high quality.3.A line drawings dataset is constructed by us,it contains the line drawings corresponding to the 2d geometric images in various student textbooks or teaching aids.The line drawings dataset contains thirteen types of line drawings.The line drawings dataset is mainly used for training and testing of convolutional neural network(CNN)parameters.4.A 3d geometric reconstruction system is designed and implemented by us.The system includes five functional modules: input image,target detection,line extraction,geometric classification and 3d reconstruction.The input image module can read the 2d geometric image and input it to the system;the object detection module can adaptively extract the geometric object in the image;the line extraction module can convert the 2D geometric object into the corresponding line drawings;The classification module classifies line drawings;the 3d reconstruction module realizes matching of line drawings and 3d geometric models,and reconstructs 3d geometric models.The system can inputting 2d geometric images and outputting corresponding 3d geometric models.