Research Of 3D Human Body Scanning And Multi-layered Garments Simulation Based On RGBD Cameras

3D body scan has been widely used in computer animation,virtual reality and fashion design.However,the conventional 3D body scanner such as laser scanner is very expensive and needs a further professional operation.RGB-Depth cameras have recently been used to capture the human body,several systems based on the depth camera have been proposed for body modeling.However,these methods usually are time-consuming and human beings cannot keep still in a long time.In the fields of textile and fashion design,3D body scanning plays a more and more important role.Hundreds of body measurements can be automatically extracted from the 3D body model in seconds.These body measurements can be further used for more application such as the garment design.However,measuring the body from 3D body model is only a simple application,and more application of 3D body scanning should be explored.In this paper,a rapid 3D human body scanning system was proposed based on 9 depth cameras,and a two-step registration method was proposed to align the point clouds captured from multiple depth cameras.With the help of the 3D scanner,we also explored more application of scanned data.The main contributions of this paper are as follows:1)A novel two-step registration method is proposed to calibrate multiple depth cameras,and a 3D body scanning system based on 9 depth cameras was proposed.The acquisition of the raw date in our system can be finished in 3 seconds.During scanning,the subject is asked to stand in the center of the system and keep still for 3 seconds.As the resolution of scanned data is higher at a closer scanning distance,and the limitation of scanning field of the single depth camera,we placed 9 depth cameras from multiple views,which can cover the whole body of the subject.Hence,each camera will capture part of the subject.Once these 9 point clouds are aligned together,the full body data can be obtained.In our two-step registration method,Zhang's calibration method is firstly used based on double-face checkerboards to estimate the position of depth cameras.This results in a coarse registration for 9 point clouds.Next,2 A4 creased papers are used to further fine registration based iterative closest point(ICP)algorithm.Due to the abundant surface details of creased papers,the 9 point clouds can be aligned precisely.This registration process is only done once,and the affine matrices of 9 depth cameras are stored.When a human body is scanned,the complex registration problem is converted to a simple affine transformation problem.Compared to other systems,rapid data acquisition and rapid registration are the advantages of our system.We also applied the multithread processingto alleviate the interference between depth cameras.2)An automatically rigging method for the scanned human body was proposed.Rigging for a body model is usually manually done by the artist using modeling software such as 3D Maya.In this paper,a rigged statistic body model is used to fit the scanned body.This results in a skinned articulated model of the scanned body.Firstly,introduced how to create the statistic body model from human body library.Secondly,we pre-rigged the statistic body model.Thirdly,the rigged statistic body model is fitted to the scanned body using an ICP-style optimization strategy.Finally,the skeleton and skin weights are transferred from the rigged statistic body model to the scanned body.3)A novel scanning-based personalized 3D mannequin reconstruction method is proposed.The resulted 3D mannequin can preserve both body measurements and body shape of the original scanned body.Currently,a 3D mannequin is usually reconstructed from body measurements,which can preserve original body measurements.These 3D mannequins can be imported to fashion CAD for garment design.However,a mannequin is usually reconstructed from body measurement or body features using current methods,which lacks accurate body shape information.For the same human body measurement,the same mannequin from is not definitive.This may result in an unfit garment for the target human body.The 3D mannequins imported into the fashion CAD should be symmetric.Therefore,the scanned body model needs further processing before it is imported into the fashion CAD.And the problem of 3D mannequin reconstruction is converted to a problem of body symmetry.When the subject stands,his/her shoulders are not absolutely the same level.So the scanned body should be adjusted to pose symmetry.We made use of the LBS to adjust the pose to be symmetric.Due to the movements of muscle and fat,the shape of the human body is also not symmetric.We further proposed a slicing method to process the scanned data.A point cloud will be obtained after our symmetric process.Finally,the symmetric mannequin surface can be reconstructed from this point cloud.Compared to conventional methods,our method can preserve not only the body measurements of the user,but also the body shape of the user.4)A scanning-based method was proposed to model the character wearing multiple layers of clothes.Despite the development of user-friendly interfaces for modeling garments and putting them onto characters,preparing a character dressed in multiple layers of garments can be very time-consuming and tedious.In our method,this is achieved by making use of real clothes and human bodies.We first scan the naked body of a subject by the 3D scanner,and a statistical body model is fit to the scanned data.This results in a skinned articulated model of the subject.The subject is then asked to put on one piece of garment after another,and the articulated body model dressed up to the previous step is fit to the newly scanned data.The new garment is segmented in a semi-automatic fashion and added as an additional layer to the multi-layer garment model.Compared to conventional methods,our method can rapidly model a character wearing multiple layers of clothes.To our best knowledge,this is the first paper to animate multiple-layer cloth based on the 3D scan.5)A novel cloth simulation method via combining LBS and physically-based simulation was proposed.To animate the dressed body,two main conventional methods can be used.The first one is real time.Binding the garment to the skeleton,and then the garment can bedeformed by deforming the skeleton.This method is simple and efficient.But it cannot work well for loose garments such as skirts.And this method cannot obtain the realistic dynamic cloth result.The other one is physically-based simulation.The naked body is animated by imported motion data,and the garment is deformed via physically-based simulation.This method can obtain realistic cloth simulation.However,to estimate the accurate parameter for physically-based simulation is still a challenging problem.Bad parameters will result in unnatural simulation results.During runtime,using the character dressed in multiple layers of clothes,we can realistically animate the clothes by blending the deformation of the clothes by linear blend skinning and physical simulation.This approach is beneficial as the shape of the cloth can be preserved in a natural fashion while physical movements of the clothes can be simulated.6)A novel method was proposed to reconstruct textured 3D textiles/garments via combining 3D scanning and texture synthesis.The 3D textile model plays an important role in textile engineering.However,not much work focus on high-quality 3D textile reconstruction based cheap depth cameras.The texture is also limited by photography methods in 3D scanning.This paper presents a novel framework of reconstructing a high-quality 3D textile model with a synthesized texture.Firstly,a pipeline of 3D textile processing is proposed to obtain a better 3D model based on Kinect Fusion.Then,convolutional neural networks(CNN)is used to synthesize a new texture.To our best knowledge,this is the first paper combining 3D textile reconstruction and texture synthesis.Experimental results show that our method can conveniently obtain high-quality 3D textile models and realistic textures.