3D Modeling Technique Of Human Body Based On Data Features

3D digital model of human being is synonymous with human avatar in Computer Graphics,and is applied in simulation scenario to represent the user herself/himself or virtual character controlled by user.Nowadays the modeling technique for human avatar is becoming a hot academic topic as the intense requirement of reality augmentation in military applications,e.g.virtual training of soldier skills,and domestic areas,e.g.filming and video games.Because of the self-occlusion of human body,discrete data capturing via hardware and other reasons,we are constrained to utilize the limited partial discrete data to build much more accurate and high-quality complete 3D model,this has brought challenge to modeling technique for human body,and therefore kept the related research continuously thriving to fulfil the increasing demand of the model quality.At present,the related works are insufficient in methodology for construction of data-capturing system and effectiveness for alignment of textured data,this remains the exploring space for our work.Additionally,the procedures of modeling and quality of final result from human modeling system would be affected by multiple technical stages,some involving factors,such as the features of human body,hardware device,texture and geometry of captured data,are worthwhile to exploit to optimize the data-processing and improve the modeling result of human avatar,these are the core and innovations of our work.Therefore this paper involves a challenging but innovative and practical research topic.This paper regards the 3D model generation of human body as a basic demand,and provides a batch of algorithms involving the construction of data-capturing system,data alignment,data rectification and so on,which targets to build the high-fidelity 3D model finally.These methods are designed or improved after careful analyses of the hardware capacities and functionality of each stage in modeling pipeline.Our work will be introduced separately in four parts.The first part is mainly about the construction procedure for data-capturing system.After analysing the characteristic of RGB-Depth camera,we provide a heuristic algorithm to set up an automatic multi-camera system for the first time with complete postprocessing mechanism,which targets to capture more precise data and cover as more surface as possible to generate 3D model for human body.In the second part,a colour-aware registration algorithm is introduced for data alignment in both rigid and non-rigid case,which will improve the alignment quality of texture and geometry simultaneously for highly-textured data.More important,the non-rigid registration part is the first method in relevant area to utilize texture to align data captured from multiple sparse angles.The next part presents the updated version of data-capturing system,each monocular RGB-Depth camera is replaced by a DIY structure light(SL)device.In this case,the brand new system is simplified in constitution and optimized in system control,meanwhile,improves the quality of the output.The final part aims to solve the problem encountered when acquiring data in hair region via SL device,a revised bilateral filtering algorithm is devised after assuming the depth smoothness locally,then this algorithm is used to repair the data and help to build hair result with better fidelity.All the algorithms presented above always focus on the characteristic of the captured data and is devised after taking comprehensive factors into consideration,such as the properties of hardware,precision and quality of the output result and systematic usability.These algorithms also cover a wide scope in modeling technique,ranging from system construction for data capturing to regional data repair.The experiment results have proved that our human modeling system has the ability to generate high-quality 3D model.In comparison with state-of-the-art works,our work makes up the deficiency in systematic designing method for data-capturing system,solves the problems about colour-aware data alignment from sparse multiple-shot and data rectification in hair region,these improvements constitute a feasible scheme for 3D modeling technique of human body which could result in more satisfactory output.