利用數位相機建構三維物體點結構

Abstract

本論文的目的在於利用相機與物體的相對運動做連續拍攝來重建出物體的三維點結構與各個視角的投影矩陣。三維點結構本身可以使用在辨識與提供物體的幾何資訊，而各視角中的投影矩陣配合物體投影的輪廓資訊可以做物體密集的幾何重建(dense reconstruction)。 文獻中對於由影像序列恢復結構與運動的方法大都要求完美的影像特徵點對應，也就是所追蹤的特徵點在所有的圖片中都不能遺失，而且所追蹤的特徵點位置沒有雜訊，然而完美的特徵點對應需求會嚴重限制實際的應用性。本論文使用了multi-view reconstruction method和iterative reweighted least square(IRLS)來做強健式的透視投影空間(projective space)中的結構與運動的重建，並利用iterative absolute dual quadric 做auto calibration來將結果校正至歐式幾何空間。本論文把此系統應用在人頭的辨識與密集重建上，初步證實了本論文所提出的系統藍圖是具有實際應用性的價值。

The current algorithms for recovering the scene structure and camera motion from an image stream require a set of well-tracked features. Such a set is in general not available in practical applications. Thus, there is a need for making the recovering structure and motion algorithm deal effectively with missing features and data noise in the tracked features. We propose a new scheme combining the multi-view reconstruction method and the iterative reweighted least square method. This scheme is able to deal effectively with the missing features and noise in the individual features. Furthermore, the proposed scheme includes an auto-calibrated method for Euclidean reconstruction using the iterative absolute dual quadric framework. The robustness of the proposed scheme is tested on both the synthetic data and the real data. For the real data, the scheme is applied to the human head recognition and the dense reconstruction of human head geometry.