View Information Based 3D Model Retrieval Method

In recent years,because of the rapid development of Internet technology,along with the increase of information dissemination speed,a large number of 3D model data begin to show spurt growth,and the data of 3D model has become a new type of data.With the advent of deep learning,the research of 3D models by domestic and foreign researchers has become a hot topic at present,and it is widely used in various industries,such as: medical 3D printing technology,computer aided design(CAD),3D animation design,industrial product design,mechanical parts manufacturing,etc.In view of the wide application prospects of 3D models,how to effectively manage a large number of3 D model data and quickly and efficiently select the required 3D model data has become an urgent need of the current environment.So far,the 3D model retrieval technology has become a technical method to solve this urgent need.This method performs the optimal model search by sorting the retrieval results according to the similarity by extracting the similarity between the models by querying the 3D model.This method does not require a large number of manual annotations,and has strong objectivity and wide applicability.This paper focuses on the multi-view based 3D model retrieval algorithm.Based on the research of a large number of domestic and foreign scholars,two new algorithms have proposed to solve the 3D model retrieval problem.(1)The 3D model retrieval algorithm based on the generative adversarial networks;(2)Another 3D model retrieval algorithm based on deep reinforcement learning.The proposed methods in this paper are based on the 3D model in the Princeton Model Net project,3D model together with the corresponding 2D image data,which are collected online,is used to establish a new 2D-3D image model database.Both algorithms are the view-based 3D model retrieval algorithm.Firstly,the 2D view sequence is extracted from the model after preprocessing the 3D model.Then the view sequence is processed by two different network frames,and the characteristics view of3 D model can be selected to represent the 3D model.Finally,the obtained characteristics view is subjected to the underlying feature extraction,and then the similarity measure between the models is performed,and the retrieval result is obtained based on that.This paper sets up a large number of comparative experiments to prove the effectiveness of the proposed search algorithm,and a new 2D-3D database is established to support the experiment.The effectiveness of the proposed algorithm is analyzed based on the experimental results.