Visual Hull Computation Based On Deformable Model

In recent years, 3D reconstruction based on image sequences becomes an very important problem in the filed of computer vision and virtual reality. The concept of Visual Hull is defined as the max volume calculated from the silhouette information of the image sequence. The recovery results of the traditional volumetric approaches can't obtain smooth surface, and the precision depends on the limitation of the memory.This thesis introduces the deformable model into the reconstruction framework. The deformable model based Visual Hull computation methods are listed as below.(1)Snake parametric model. On the basis of snake deformable model, the algorithm translates the visual hull computation problem into a forces computation problem. The initial mesh is deformed under the internal and silhouette forces until convergence from a sphere to the visual hull. This method uses the images sequences as input, and computes the internal and external forces from the smoothness and silhouette constrain.(2)Level set geometry model. Although the Snake model can obtain the smooth results, it can't easily change the topology of the deforming surface. Level Set algorithm embeds the deform surface into uniform grids, and it can track the deformation of the surface implicitly. The most important advantage of Level set algorithm that it can change the topology of the surface freely.In the experiment, we design our own hardware platform for collecting the image data. Our deformable approaches reduce the mesh mistake; improve the quality of the mesh; and recover the object robustly. The experimental results show that the final mesh has higher precision and smoothness than the traditional schemes.