Font Size: a A A

Research On Generating 3D Human Avatar With Texture Map Based On Images

Posted on:2021-03-03Degree:MasterType:Thesis
Country:ChinaCandidate:Q HanFull Text:PDF
GTID:2428330614460339Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the rapid development of computer simulation technology,people's life is becoming more and more rich.At the same time,it makes them more and more pursue the realistic experience and visual enjoyment brought by technology.In the large-scale crowd simulation scene,virtual try on system,movie animation,video conference and game etc,the realistic and refined 3D character model brings different visual impact to people.In this dissertation,we study how to automatically generate a visualized 3D character model from a character image.Firstly,we use the method of deep learning to extract the semantic feature information of the character in the image,and then map the resulting human body analysis map to the texture space to generate part of the texture,and then get a complete UV texture map through the texture stiching technology;secondly,we integrate it into the standard mesh model SMPL to get the target 3D mesh model.The UV texture map is mapped to the corresponding 3D character model through the texture coordinates,so as to achieve the creation of the avatar of the target character.The main work of this dissertation is as follows:1)The current research status and achievements of the visualization of three-dimensional personal model from images are reviewed.Summarizing and improving the systems and algorithms currently used in the generation of three-dimensional persona model from image to visualization,and the network model used to extract the semantic feature information of persona from image.2)In view of the lack of human body details and the low accuracy of semantic tag map in the existing human body parsing methods,a cross refinement network(CRN)is proposed,which combines boundary and semantic information.The semantic feature information of the characters is extracted from the images for subsequent operation.3)A complete UV texture is obtained by combining the semantic feature map with some texture stitching.Then,the semantic feature map is fused into the standard pose model SMPL to generate the target 3D mesh model.Finally,the UV texture coordinates are written into the obj model file to get the personalized 3D human model.
Keywords/Search Tags:3D character model, semantic segmentation, human parsing, UV texture, Cross Refinement Network
PDF Full Text Request
Related items