Research And Implementation Of A Real-time Virtual Try-on System

With the development of e-commerce and the growing demand for virtual experience,the research on virtual fitting system based on cloth simulation technology and virtual interaction technology has received extensive attention.The interactive virtual try-on mode gives users a new shopping experience.The virtual fitting system based on 3D cloth simulation has become a hot topic in current research.The biggest bottleneck of the current virtual fitting system is that the simulation speed of the clothes is too slow to synchronize with the user's interaction,that is,the real-time cloth simulation system cannot be realized.With the continuous improvement of the computational power of graphics processing units(GPUs),its application in cloth simulation has also become a research hotspot.This paper proposes a real-time simulation algorithm based on GPU,which maps all data and computations to the stream and kernel of the GPU to make full use of the parallel computing power of the GPU.The algorithm mainly includes static cloth simulation and dynamic cloth collision processing.Mass-spring system is used to model cloth,which can characterize the characteristics of the clothes well.The human body uses the SMPL model to characterize the details of the human body.In the static cloth simulation,Verlet integration is used for position update,then the static human body is used as the collision body.Finally,we use a method based on position mapping and inelastic collision to deal with static collision.Because Verlet integration has good stability,we can take large steps for time integration,which speeds up the simulation of cloth.In the dynamic cloth collision processing,the collision is handled by the swept space of the human body between adjacent frames,and we use the local driving method to deal with the dynamic collision.Since collision detection is the most time-consuming in the whole cloth simulation,this paper uses BVH(Bounding volume hierarchy)to accelerate detection.We implement a GPU-based BVH parallel construction method,which uses Morton code to encode collision elements.Then,a binary radix tree(BRT)is constructed in parallel to create a hierarchy,and finally the bounding box of each internal node is updated in parallel.S-BVH and D-BVH were constructed for static and dynamic collision detection acceleration,respectively.In order to solve the failing collision detections in the static simulation caused by the thin bounding box,we extend the bounding box in the normal direction of the corresponding triangular surface,and then the extended bounding box is used for collision detection,which greatly reduces the detection failure situations.Experimental results show that our method has faster speed than the previous related technology,while retaining the realistic effect of the fabric.Finally,we implemented a real-time virtual fitting system based on the algorithm.The system consists of three modules,Kinect human skeleton tracking,fast cloth simulation algorithm and OpenGL-based rendering module.The system supports basic clothing selection,360° clothing display.In addition,the 3D point picking algorithm is used to realize real-time interaction between the user and the system.The user can drag and drop the simulated clothes to simulate the daily fitting behavior.