【摘要】 针对各种传统可视外壳生成算法中数据冗余及壮健性不足等问题 ,提出了一种新的从阴影图像中快速重构物体可视外壳的壮健的算法 ,即首先利用物体表面的拓扑结构直接生成外壳 ,然后使用改进的 Surface Net算法光滑三维表面 ,从而在保留经典的体求交方法壮健性的基础上 ,克服了对于物体内部点的冗余计算和存储问题 ,不仅使得计算的时间复杂度降低到仅线性依赖于外壳上结点的数目 ,而且降低了像片数对算法复杂度产生的影响 ,实验结果表明 ,在算法复杂度和壮健性上优于诸如八叉树等传统可视外壳算法。更多还原

【Abstract】 The traditional volumetric visual hull generating methods were not applicable to huge-volume objects due to redundant calculations. Other methods based on ray intersections were sensitive to input perturbations and were hence lack of robustness. A fast new algorithm was represented in this paper for reconstructing an object’s visual hull from its silhouette. The topological structure of the object’s surface was taken into consideration, and the surface mesh was reconstructed directly without having to compute the redundant information of all the voxels inside. Then, an improved SurfaceNet algorithm was adopted to smooth the 3d surface. The robustness of the classical volume carving method was reserved, while the time complexity was reduced to only linearly depend on the number of vertices on the final surface. The dependence of the time complexity on the number of photos was reduced as well. The results of the experiments show that this topological generation algorithm is superior to some classical visual hull methods, as far as its reconstruction function for practicality is concerned.更多还原