3D Virtual Try-on Based On Kinect Motion Capture

In recent years, with the development of the natural human computer interaction,extensive attentions and researches have been carried out in virtual fitting room, which isbased on the body feeling interactive techniques. Different with traditional virtual fittingrooms, it is very convenient and natural for customers to try-on clothes before purchasing inthese rooms, which will bring them a whole new user experience.In current market, those practical virtual fitting rooms focus on the style of clothing.Without the interface to intuitively simulate natural collision attributes between the virtualroles and the clothes, these fitting rooms are still deficient. The main goal of this thesis is todevelop the presentation of the gesture with virtual character, which can detect the collisionbetween human body and clothes in real time. Then we build a bridge between the virtual andreal world via a posture recognition system, which will increase the fun of fitting for users.This paper builds a virtual fitting platform based on Kinect including four sections.Firstly, we use Forward Kinematics to calculate the deformation matrices of skeleton bonesby referring to the analysis of the motion capture data BVH file, then convert the movementsto hierarchically same characters. And based on these operations, customer can try-ondifferent garments using the system to recognize the human action by detecting the handgesture. Secondly, according to the hierarchical structure of the human body skeleton, somepreprocessing operations on the human body and clothes are conducted: we construct thebounding boxes and assigning some extra geometric constraints to the garments, and then usethe physics engine PhysX and APEX dynamic framework to simulate the clothes.Furthermore, in order to achieve realistic results, we build a virtual scene with some aideddesign tools. Besides, we implement some skinning algorithms on GPU so as to acceleratecomputing and ensure the real-time. Experimental results show that the platform is real-time,reliability and authenticity.