提出一种对可变性物体表面的动态变形过程进行三维重建的方法。将图像序列第一帧中物体表面划分成多个图像块,利用优化函数寻找每个图像块在所有帧中的对应关系,最终实现物体运动过程的三维重建;完成每一帧的三维重建后,根据图像关联度和局部拓扑结构对结果进行错误检测,并对出错块进行重新优化。模拟实验、实际数据实验以及与同类算法的比较,验证了该方法的有效性和可行性。
This paper presented a method of 3D reconstruction for dynamic deformable surfaces. This method first divided the surface in the first frame into lots of small patches, then optimized the correspondences for all patches in every frame, and finally finished the 3D reconstruction for the dynamic surface. After reconstructing one frame, it performed error detection according to image NCC and local topology, and re-optimized erroneous patches to improve the robustness. The performance and effectiveness were evaluated by simulated and real data experiments.