使用共線攝影機系統及景深範圍重建三維場景之技術

Abstract

本論文針對多基線立體提出以影像景深範圍資訊降低原先多攝影機特徵點比對的複雜度，同時也提昇特徵點比對的正確性。有了比較好的特徵點比對結果，即可得到更好的三維重建結果。 此外，我們將多基線攝影機系統以基線上一點為中心旋轉，每隔固定旋轉角度間隔拍攝一組環境影像，一方面整合各組影像之三維重建資訊成為更完整的重建結果，另一方面對不同組影像間的場景重疊部分做交互檢查，算出更穩定正確的結果。 利用三維環場的重建結果，可以變動視點計算出新的場景影像。 我們將以實驗來驗証我們的方法。

This thesis presents a method of using depth of field to reduce the computational complexity of multi-baseline stereo matching and to raise the correctness rate of feature matching at the same time. With the improved feature matching result, we can reconstruct a better 3D panoramic model of the scene. Next, let the multi-baseline camera set be rotated around one endpoint of the baseline to shoot a sequence of images of the scene in an increment of a fixed rotation angle. On the one hand more complete panoramic model of the scene can be reconstructed by integrating the 3D reconstruction results obtained at each shoot, and on the other hand more robust reconstruction is derived by executing the cross-reference checking on the reconstructed parts which are redundant during reconstruction. New images can be generated at different viewpoints with the 3D panoramic scene reconstruction. Experimental results are provided to demonstrate the proposed method.