This thesis provides a complete framework that enables the creation of photorealistic 3D human models in real-world environments. The approach allows a non-expert user to use any digital capture device to obtain four images of an individual and create a personalised 3D model, for multimedia applications. To achieve this, it is necessary that the system is automatic and that the reconstruction process is flexible to account for information that is not available or incorrectly captured. In this approach the individual is automatically extracted from the environment using constrained active B-spline templates that are scaled and automatically initialised using only image information. These templates incorporate the energy minimising framework for Active Contour Models, providing a suitable and flexible method to deal with the adjustments in pose an individual can adopt. The final states o f the templates describe the individual’s shape. The contours in each view are combined to form a 3D B-spline surface that characterises an individual’s maximal silhouette equivalent. ududThe surface provides a mould that contains sufficient information to allow for the active deformation of an underlying generic human model. This modelling approach is performed using a novel technique that evolves active-meshes to 3D for deforming the underlying human model, while adaptively constraining it to preserve its existing structure. The active-mesh approach incorporates internal constraints that maintain the structural relationship of the vertices of the human model, while external forces deform the model congruous to the 3D surface mould. The strength of the internal constraints can be reduced to allow the model to adopt the exact shape o f the bounding volume or strengthened to preserve the internal structure, particularly in areas of high detail. This novel implementation provides a uniform framework that can be simply and automatically applied to the entire human model.
展开▼
机译：本文提供了一个完整的框架，该框架可以在真实环境中创建逼真的3D人体模型。该方法允许非专业用户使用任何数字捕获设备来获取个人的四张图像并创建个性化3D模型，以用于多媒体应用。为此，系统必须是自动的，并且重建过程必须灵活以解决不可用或错误捕获的信息。在这种方法中，使用受约束的活动B样条模板自动从环境中提取个人，该模板仅使用图像信息进行缩放并自动初始化。这些模板结合了用于主动轮廓模型的能量最小化框架，从而提供了一种合适且灵活的方法来处理个人可以采用的姿势调整。模板的最终状态描述了个人的形状。每个视图中的轮廓线都合并在一起，形成了一个3D B样条曲线表面，该表面表征了一个人的最大轮廓等效值。 ud ud表面提供了一个包含足够信息的模具，以允许基础通用人体模型的主动变形。这种建模方法是使用一种新颖的技术执行的，该技术将主动网格演化为3D以使基础人体模型变形，同时自适应地约束它以保留其现有结构。主动网格方法结合了内部约束条件，这些约束条件可以保持人体模型顶点的结构关系，而外力会使与3D表面模具相接的模型变形。可以减小内部约束的强度，以允许模型采用边界体积的精确形状，或者可以增强模型以保留内部结构，尤其是在高细节区域。这种新颖的实现提供了一个统一的框架，该框架可以简单，自动地应用于整个人体模型。
展开▼
