室內環境之三維模型重建

Abstract

從二維的影像資訊重建出三維的場景模型，一直是電腦視覺領域一個重要的 研究主題，隨著電腦計算速度的進步，這項研究所延伸而出的應用更是琳瑯滿 目。近年來蓬勃發展的電腦繪圖、虛擬實境等等，都會利用到影像重建的技術， 比如將一些現成的玩具利用多個視角的照片，便可在電腦中產生玩具的模型。我 們提出了一個透過影像來重建三維場景模型的方法，透過影像之中與影像之間的 關係將攝影機的內外部參數算出之後，我們變能夠將影像間重複拍攝的部份的三 維座標點算出。在算出三維座標點之後，使用Wendland將三維座標點之間缺乏的 部份算出來形成場景的三維表面模型並將拍攝到的影像當作場景的材質貼到該 模型上以達到擬真的室內環境重建。

In recent years computer hardware and computer graphics has made tremendous progress in visualizing 3D models of real objects. Many techniques have reached maturity and are being ported to hardware. This seems like in the area of 3D visualization, performance may increase even faster than Moor’s law. Some job required a million dollar computer a few years ago can be now achieved by a custom computer, which cost a few hundred dollars. It is now possible to visualize complex 3D scenes in real time due to the advancement of computer hardware. This speed of evolution causes an essential demand for more complex and realistic models. Even though we are now able to build three-dimensional models, the tools for three-dimensional modeling are getting more and more powerful, synthesizing realistic models is difficult and time-consuming. Many virtual objects are inspired by real objects, so we are interested in being able to build three-dimensional environment models directly from the real environments. In the past, visual inspection and robot guidance were the main applications. We require more and more 3D content for computer graphics, virtual reality and communication nowadays. The visual quality becomes one of the main points of attention. Therefore not only the position of a small number of points have to be measured with high accuracy, but the geometry and appearance of all points of the surface have to be measured. We proposed a semi-automatic 3D indoor environment reconstruction procedure using the thin-plate splines for surface modeling and texture mapping. First, the intrinsic parameters of the two cameras are calibrated. Second, calculate the fundamental matrix by using the well-known Eight-Point algorithm and the essential matrix is derived to be the combination of fundamental matrix and the two camera intrinsic matrices. Third, relative pose of the two cameras can be extracted from the essential matrix and sparse 3D point reconstruction can be performed. Forth, interpolate 3D surfaces among the reconstructed sparse 3D points with the thin-plate splines. Finally, we can add textures on the reconstructed 3D surface model with some texture mapping techniques. The 3D surface model established i with the proposed reconstruction system provides useful information for robot navigation and other applications.