基于时空关联块匹配的动态变形表面三维重建

3D reconstruction of dynamic deformable surface based on spatio-temporal patch matching

静态物体三维重建已经得到广泛应用,但是动态变形物体的三维重建仍然存在问题.基于块匹配的思想,提出一种时空关联的动态变形表面三维重建的框架,将图像中一个物体表面划分成多个小块,独立处理每个块,用一个时空关联的优化函数进行块匹配.优化函数中,同时考虑了立体图像序列中四个对应的图像块.其中左、右图像块之间进行仿射变换,前、后图像块之间进行平移变换.模拟实验、实际数据实验以及与同类算法的比较,验证了该框架的有效性和可行性.

3D reconstruction of static object has been applied to many fields.However,3D reconstruction of dynamic deformable object is still tough to solve.This paper presents a novel framework for reconstructing dense 3D dynamic deformable surface with spatio-temporal coherence by patch matching with calibrated stereo image sequences.A surface can be divided into lots of small patches,and every patch can be dealt with independently.After dividing,the framework is implemented by a designed spatio-temporally coherent energy function.Simultaneously,four patches in stereo image sequence are taken into account.The performance and effectiveness are evaluated by simulated and real data experiments.