Real-time 3D Clothing Dynamic Virtual Try-on Experience System Based On Somatosensory Measurement

With the improvement of people’s life,the modern fashion culture with high-quality and personalization has promoted the rapid development of clothing customization industry.Recently,crossing with the new generation of information technologies,such as 5G,AI,Io T and so on,,the virtual clothing try-on technology has become research highlights at home and abroad.Specially,the interactive person-by-person clothing virtual experience can illustrate digitally the front-end demand of fashion customization industry with "human-centered" tendency.Based on the principles of 3D vision sensing theory,clothing modeling and the modeling tools such as Unity3 d and Maya,this paper solve the problems of the dynamic 3D virtual try-on system featured by the personalized postures tracking,real-time interaction and low cost.The research subjects focused on the extraction of Kinect’s human body bone features,perception measurement of body circumference size,3D dynamic-feature modeling and tracking of clothing,real-time display of virtual try-on for a real persons.The main innovative contributions of this paper have been given as follows:Firstly,by using Kinect’s data,a pixel clustering method based on local feature’s similarity was proposed to extract the human profile features with the optimal extraction efficiency.Then,the GBDT(Gradient Boosting Decision Tree)calibration model of the body circumference was designed to fit the circumference size of the human body with improved fitting accuracy.Secondly,a 3D clothing perception model based on the dynamic feature points was constructed to percept body postures.Fusing with the calibrated body circumference sizes,the perception model can match the profile of human body precisely.Furthermore,In order to distinguish body motion features,a GL-SVD(Global-Local Singular Value Decomposition)algorithm based on hierarchical control was designed to accomplish 3D coordinative mapping between real person and clothing model,resulting in a real-time cooperative tracking and high-precision matching.Finally,in the development environment including Unity3 d,Maya and Kinect,we have completed the architecture design,module development and integration of the 3D dynamic virtual try-on experience test system.The test system includes data acquisition process,3D clothing perception modeling,dynamic clothing try-on,real-time interactive control,virtual display,with result of a complete 3D dynamic virtual try-on experience test system(software).In Windows 10 operating system,this paper integrated the test system software and Kinect tool to build a virtual try-on interactive test environment.The try-on comparison test cases include men and women,typical body postures of such as side,akimbo and leg lift as well as blouse,pants and skirt,etc.The test results showed that the improvement on tracking precision of GL-SVD method is 10%,the mean error(MAE)of GBDT model is 4.3,and the root mean square error(RMSE)is 4.6.This indicates that the proposed 3D clothing perception model had high tracking accuracy with human body shape.Compared with the reported literature,the proposed system had fit diverse clothing types and body posture types with a strong experience of reality.In summary,the 3D dynamic virtual try-on experience test system can complete the,dynamic clothing-person tracking with high precision in real-time.The system can adapt to the personalized features of human body and reach a "try while moving" experience.