由三張無校正物體的影像估測相機的外部參數

Abstract

在本篇論文中，將提出一以三張有兩個共同可視平面的影像來線上校正（on-line calibration）相機外部參數的方法。這個方法希望可以讓使用者透過手持的定焦相機拍照並環物重建三度空間物體模型。首先估測出相機外部參數，再與先前已校正的相機內部參數結合以求出相機投影矩陣。透過影像間的投影關係便可估算出物體上各點的三度空間座標，並將所有的三度空間座標資料轉換到同一座標系的三度空間中。最後便可以數個粗略的大平面逼近物體形狀以完成重建。在本文中也將會提出以虛擬物體及實際物體拍攝影像所得到的實驗結果來印證本方法的可行性。

This thesis presents a linear on-line extrinsic camera calibration method by taking three pictures of a scene containing two commonly visible planar regions. Based on this method, the user can use a hand-held, fixed-focus camera to reconstruct the three dimensional object. First, we find the extrinsic camera parameters, and then we combine them with the intrinsic camera parameters, which were determined in advance, to derive the camera projection matrices. The 3D coordinates of object points can be computed by triangulation based on the derived projection matrices. It is also necessary to transfer all the 3D data into the same 3D coordinate system through the coordinate transformation. Finally, we use several planes reconstructed to approximate the shape of the object under consideration. Experimental results with synthetic and real pictures are provided to demonstrate the proposed method.