單一視角影片中折疊表面之深度重建

Abstract

這篇論文提供了新的方法還原影片中有皺折的表面的深度值, 像是衣服或飄動的旗子, 並使其結果可應用在立體顯示器上. 我們的方法結合了Shape-from-shading和基於主軸分析的子空間近似法, 使結果能保留表面的高低起伏變化, 同時不會受影片中的雜訊影響.

In this thesis, we propose a novel method to reconstruct fluttering surface such as flags and cloth in the video sequence which can be used for the 3D display. Shape recovery of real object from a video sequence is a difficult subject. Here, we focus only on fluttering surface which can possibly be folded. While most shape-from-shading can only deal with single-material smooth objects, we propose using shape-from-shading and decoloring techniques to reconstruct more detailed surfaces under a single directional lighting condition, the surface can be multi-material with folding. To alleviate the noise and ill-pose problem, we take shape-from-shading as initial-guess, and further use Principle Component Analysis (PCA)-based subspace approximation to recover full video sequence.