Research On 3D Human Body Modeling Based On 2D Point Cloud

In some related areas such as virtual fitting,three-dimensional(3D)games,3D movies and so on,all of them are inseparable from the research of 3D digital modeling of human body.Especially in costume design,costume order,virtual fitting and other industries,the accuracy of 3D modeling directly affects the effect of the later period.In these applications,we need to measure the 3D human body model,obtaining the size information of the human body and other operations,so as to conduct virtual inspection and inspection on the clothing before the formal upper garment,so as to timely find problems and improve them,shorten the production cycle,and reduce unnecessary cost waste.Therefore,it is of great significance to study how to construct a virtual 3D human body model that is similar to the real human body shape quickly and realistically.The main research work is as follows:Firstly,in order to increase the diversity of human body morphology in the 3D human body model library,a convenient 3D human body data enhancement technology was introduced,which firstly trained a neural network from the proportion of an important part of the human body to the PCA.and then quickly generated the 3D human body model by editing the proportion of human body parts.The enhanced human body model library was classified and processed,and representative 3D human body model was selected as the data source of the experiment.Secondly,a new 3D human body modeling method is proposed,This method can not only deal with two-dimensional(2D)point cloud with uniform sampling,but also deal with 2D missing point cloud.In traditional methods,3D human body reconstruction based on point cloud usually takes 3D point cloud as research object.However,this thesis proposes the rapid reconstruction of 3D human body model based on the 2D point cloud for the first time.At the same time,considering that it is a challenging problem to reconstruct a complete 3D human mesh model from an incomplete 3D point cloud,the method in this thesis simulates the damaged and incomplete state of the 2D point cloud by cracking the 2D point cloud,so that the algorithm can process the incomplete 2D point cloud.The main content includes:The 2D point cloud is fractured by correlation method,and then build a data set composed of 2D point cloud image and corresponding human body black-and-white binary image,and train a generative adversarial networks model generated by the former to generate the latter.The model converts the 2D point cloud image into the corresponding black-and-white binary image.The binary graph is input into a trained convolutional neural network to evaluate the effect of constructing a 2D image into a 3D model.A large number of experimental results show that the 3D human body model reconstructed by the method in this thesis can effectively realize the visual sense of reality.In order to quantitatively analyze the accuracy after reconstruction,the most representative human body characteristics of waist and chest are selected as the error evaluation.At the same time,in order to further highlight the effectiveness of this method for reconstruction of 3D model based on the missing point cloud,a comparative experiment is introduced,and this method is compared with the traditional surface reconstruction algorithm.The experimental results show that the effect of reconstruction of 3D human body model based on 2D missing point cloud is better than that based on incomplete 3D point cloud.