The Application Of 3D Human Reconstruction And Virtual Dressing Based On Point Clouds

3D reconstruction is one of the most important research branches in the field of 3D vision.With the rapid development of high-performance depth-sensing equipment and technologies,the field of 3D reconstruction has attracted more and more attention from the academia.It has broad application prospects in some new fields,such as autonomous driving and virtual dressing.Virtual dressing is a kind of technology that has emerged with e-commerce booming in recent years.Consumers can know about the dressing result by dressing clothes on the virtual model directly to obtain a more reliable shopping experience.In this paper,depth-sensing cameras are used to scan users.Based on the point clouds obtained,an application of 3D human reconstruction and virtual dressing has been developed,and on this basis,a dressing system has been built.The dressing system aims to minimize the interaction with users,and a higher flexibility than old processes.In this way,it can make the system easier for users without reducing the efficiency.The specific work is as follows:First,the research of 3D human reconstruction has been carried out.Point clouds obtained by the depth-sensing cameras are registered by running a registration algorithm,and by using the RGB information of the points,the accuracy of registration is improved.Point clouds of the complete scene will be used to extract the point clouds of the human body.By running a surface reconstruction algorithm,the construction of the 3D human will be finished.The human model can reproduce the physical characteristics of the users.It will make the expression of the system more natural.Secondly,the research of clothing information extraction has been carried out.With the help of the object detection network,the clothing image input by the users is processed to output its target category,bounding box and pixel area.These contents will be seen as the texture information to serve the process of the 3D clothing reconstruction.The clothing image in a correct pose input into the module can be arbitrarily specified by the user,thus ensuring that the system has a higher versatility.Finally,the research of virtual dressing and displaying has been carried out.According to the human model and the texture information provided by the previous process,the garment pieces are instantiated based on its template preset in the system.The garment pieces will be stitched around the body.By using collision detection,the garment model will naturally attach to the body to generate the final result of dressing.It will be output to a screen in using of visualization,and finally displayed to the users.This process will be completely separated from the intervention of the users,so it can lower the threshold for the users who want to use this system.