3D Fitting Research Based On Human Body Segmentation And Deformation

In recent years, e-commerce has been vigorously developed. Users are more inclined to online shopping. Buying clothes online is more and more popular. But users cannot actually try it on, so the users demand for virtual try on more and more urgently. In this paper, we get the parameters of the human body through the automatic segmentation algorithm. According to these parameters, the standard human model is deformed to generate the specific body model. At last, dress models are matched with the specific model by bounding boxes and synchronous deformation. The main research contents are as follows:(1) Automatic segmentation of scanned body model and standard body model. By introducing the curvedness parameter, the characteristic edges are defined by the curvedness of the vertices. And a new automatic segmentation algorithm based on characteristic edge is proposed, which effectively avoids the over segmentation problem of the region growing algorithm.(2) Deformation of the standard body model. In order to realize the body parameter control, we divide the deformation into the surrounding deformation and the height deformation. Waist line is used as the source of constraint, and the length of the waist line is controlled by constraint deformation. As to the height deformation, we choose the free-form deformation algorithm. Bounding box is used to realize free-form deformation, so we can control the length of the box to achieve the height parameter control.(3) Clothing matching of deformed human body. A clothing matching method based on bounding box is proposed. The point sets for deformation of the garment model are obtained by the bounding boxes of the body model. Then, we can get the mapping of the deformation area of the two models through eliminating the redundant points. According to this mapping, the matching of clothing and human body is achieved.The fitting system constructed by this way is convenient and intuitive, and we don’t need to reconstruct the clothing models of different users. Thus, we can greatly shorten the matching time of clothing in the application.