Creation Of 3D Human Model

The creation of 3D human model is a classical researching topic in computer graphics, which building the 3D geometric human model in the virtual scene by using the computer. The topic is very challenging mainly due to two factors: the complexity of the 3D human body, the easy detection of the model distortion.However, it is not convenient for a general user to build desirable results by using the traditional 3D creation approaches, such as the 3D creation software and 3D scanner. For the 3D creation software, the user could use it to build the model in the manual way. But the operation of the software is very time consuming, and only skillful artists have shown the creation capbilities. With the development of 3D laser scanner, it is feasible to reconstruct the 3d models by directly scanning the subject. However, it has been only used in research or industry field, since the 3D laser scanner is very expensive. The drawbacks of the above two approaches have limited the development of 3d human modeling. The general users still can’t generate the 3d human model in the easy and quick way.Fortunately, the commodity depth camera has developed in the recent years. The cheap scanner device is affordable for a general user. But, the capture accuracy of the raw depth data is so low that it could not satisfy the 3D reconstruction level. This fact puts new challenge to the processing of raw depth data.By using the new-type depth camera device(e.g. Kinect), the paper studied the creation of 3D human model. Based on the observation that the human model is determined by the pose and shape attributes, the subspace of the human model is further utilized to improve the creation efficiency. In a short, the major contributions are listed as follows:1. We propose a new third order graph matching algorithm and hole filling algorithm based on Bézier curve, then present a system for fast capture of personalized 3D avatar using two Kinects. The capturing process can be finished in a moment. The system can be set up easily and used outdoor which means it has great application value. Since the capturing process can be finished in a moment, the person being captured can hold a static pose stably and comfortably. This fast capture is achieved by using two calibrated Kinects to capture the front and back side of the person simultaneously. To alleviate the view angle limit, the two Kinects are driven by their automatic motors to capture three scans covering the upper, middle and lower part of the person from front and back respectively, resulting in three partial scans for each Kinect. After denoising, all partial scans are rigidly aligned together using a novel supersymmetric third-order graph matching algorithm. Since all these partial scans can be captured in a moment, the discrepancy between them caused by body movement is neglectable, saving the effort of non-rigid alignment. The missing gaps between the front and back scans are filled using quadratic Bézier curve. The final reconstructed mesh model demonstrates good fidelity against the person with personalized details of hairstyle, face, and salient cloth wrinkles.2. We propose a parametric human model creation algorithm based on human subspace and single frame data captured by depth camera. It gets over the limitation of accuracy of depth camera and capture condition. Then, we build a system called “Instant capture”. Based on the depth image captured by single commodity depth camera Kinect and the human model subspace, we can generate the 3d human shape similar to the real scan data. Different to the traditional human modeling methods: there is a lot of incompleteness and noises in the single depth image because of the lower accuracy and capture condition, so it is difficult to utilize the local geometry feature to reconstruct accurate models. Observing the prior knowledge that human can be seen as a subspace combined by shape and pose attribution, we can complete the weakness of the raw data. The experiment has demonstrated that we can generate reasonable human shape based on the single depth image captured by the Kinect.3. We propose a markerless method for realtime reconstruction of an animating human body baesd on improved realtime SCAPE model, in which a sequence of deforming meshes is reconstructed to represent a given performance captured by a single commodity depth camera. We achieve realtime single-view mesh completion by enhancing the parametric SCAPE model. Our method, which we call realtime SCAPE, performs full-body reconstruction without the use of markers. In realtime SCAPE, estimations of body shape and pose parameters, needed for reconstruction, are decoupled. Intrinsic body shape is first precomputed for a given subject, by determining shape parameters with the aid of a human database. Subsequently, per-frame pose parameter estimation is performed by means of linear blending skinning(LBS); the problem is decomposed into separately finding skinning weights and transformations. Again, the skinning weights are determined offline from the human database, reducing online reconstruction to simply finding the transformations for LBS. This is formulated as a linear variational problem; carefully designed constraints are used to impose temporal coherence and alleviate artifacts. Experiments demonstrate that our method can produce full-body mesh sequences with high fidelity.4. We propose a parametric editing algorithm based on revised SCAPE model for a human body. Easy editing of a 3D dressed human avatar is central to many practical applications. However, it is easy to produce implausible, unnatural looking results, since subtle reshaping or pose alteration of avatars requires global consistency and agreement with human anatomy. By using the former revised SCAPE model, we show that the parameters of the model can be estimated directly from a dressed avatar, and that it can be used as a basis for realistic, real-time editing of the dressed avatar mesh via a novel 3D body-aware warping scheme. This allows the avatar to be easily controlled by a few semantically meaningful parameters, e.g., height, weight, body joints, etc. Our experiments demonstrate that the practical system can interactively produce visually pleasing results, even for a subject dressed in loose cloth.