Measurement And Editing Of 3D Human Body Based On Neural Network

As the games,animation and film industries enter the 3D era,major companies have been constantly pursuing the realistic effects.Taking the largescale action-type single-player games as an example,the 3D effect of the game is closer to reality,and the more susceptible it is.The popularity of the game is inseparable from the characters,because rich characters can increase the playability of the game.To build these different characters,you need a large number of 3D human body models as a support,so now there is more and more research work on 3D human modeling.In addition,from the perspectives of clothing,virtual fitting,personalized human body 3D printing,etc.,the application of 3D human body modeling technology is becoming more and more practical.This paper mainly explores the dimensional measurement and shape editing of 3D human body model,and proposes to calculate the size data of the human body model through the barycentric coordinate system,and further explore how to fit the circumference from the width and thickness.After that,a neural network-based human body model editing algorithm was established,which drives the deformation of the human body model by editing different attributes,and did a lot of attribute testing work for this purpose.The main research work of this paper is as follows:First,we select the marker points in the part to be measured of the template,calculate the barycentric coordinates of each marker point in turn to establish the barycentric coordinate system,and obtain relevant data of the waist and chest of all the human body models in the database.After that,we further explore whether the width and thickness of the above two human body parts can predict the circumference.We establish a fitting function based on linear regression and neural network respectively and the fitting result and the circumference is measured by the barycentric coordinate system method.Combined with the real human body data verification accuracy,the results show that the fitting function has the same effect on the real human body,and the error is within the allowable range,which indirectly illustrates the feasibility and effectiveness of the barycentric coordinate system for measuring the 3D human body model.Secondly,the barycentric coordinate system is used to obtain the skeleton points of the human body model.The pre-defined 30 body attributes are calculated from the skeleton points,and a 3D human body model editing algorithm based on the neural network is established.Using the visualization technology to compare the predicted new model with the original model in the database,it is found that the difference between the two models is small.The error between the same index vertex and the execution time of the algorithm can show that our algorithm performs well.Finally,the multi-round test is carried out on the original 30 attributes.Each round is based on a large number of human body models.The fuzzy test is used to screen out the attributes with good performance.Perform a series of tests on each property under reservation to improve the attribute interval.We obtained a number of key attributes under the human body model editing algorithm by changing multiple attributes to generate new models,and the feasibility and effectiveness of editing the 3D human body model based on neural network are proved.