3D Human Body Morphing Based On The Correspondence Of Feature Points

The method of 3D human body modeling which has developed for decades of years has a wide range of applications. As a computer graphic technique used in garment industry 3D human models are mainly applied for automatic measurement,3D virtual try on system and animation simulation, especially when these applications are used for custom tailor system. Therefore studying the method of automatic 3D human body modeling has extensive value in research and prospect for application. In this thesis, the method of deformation for 3D human body modeling is developed using 3D surface range scanned data. Three critical techniques follows.First, several current methods of obtaining 3D human body surface information are discussed. The non-contact method based on 3D body scanning technology is proposed for its effectiveness and quickness. According to the traits and the feasibility of automatic body measurement system and 3D virtual try on system, a template-based approach for 3D human body modeling is proposed. The required body model can be deformed from a template model when a number of feature correspondences between template and scanned range data are built.Secondly, a 2D shape feature representation called spin image is discussed by relating the correspondences of features between instance models and a template model. Spin image based on an oriented point compress the dense data of 3D points to a 2D image. This method facilitates the process of matching.29 landmarks are pre-marked on the surface of a male body as our template manually. For each landmark, its corresponding vertex on the instance model will be regarded as the landmark when its similarity measure is of the maximum value. The algorithm is verified on scanned human bodies with various shapes. The results show that most landmarks are found accurately. Furthermore, the implementations of using our method in other extensive applications are demonstrated, such as the body segmentation and the extraction of skeleton system.For the third critical technique, we demonstrate the deformation of the template model with the detected 29 correspondences between template and instance model to acquire the result using pyramid coordinates, which indeed is a local shape representation. The pyramid coordinate capture the local shape of the mesh around each vertex and help maintain this shape under deformation. They are based on a set of angles and lengths relating a vertex to its immediate neighbors. Results show natural looking deformations and morphing sequences with user interaction can be acquired.Finally we discuss the future research direction of feature-based deformation technical and draw the conclusion. Our algorithm is fast and simple to implement for 3D human body modeling from scanned data. Therefore, good foundation is settled for developing dynamic dressing clothes later.