環場全景製作技術及其在虛擬實境應用

Abstract

本論文的主要目標是探討環場全景影像的製作技術，若使用者事先在一 真實的環境中挑選一觀察點，並在觀察點附近架上一相機，並以相機中心 點為旋轉軸且對著360度的四周環場去拍攝多張(24張)影像的話，我們可 透過一些影像處理程序，如歪斜影像的修正(tilted image correction) 、影像彎曲(image warping)、對連續兩張影像做影像縫合(image stitching)並調整其影像之間的亮度差異(image blending)等，之後希望 將這二十四張影像圍成一環狀圓柱體。而這其間的製作程序我們希望可以 提供一個自動化的環場全景影像製作。 若相機是由使用者手拿拍攝影 像的話，可能會造成拍攝影像時出現幾張影像有較嚴重的歪斜或仰俯角等 問題，此時我們可以利用2D投影幾何轉換(2D projectivetransformation)的理論，再介入人工的方式來選取其影像之 間的特徵對應點，也希望對拍攝較不佳的影像加以修正，並仍可製作出一 環場全景影像。 最後，我們再做一些環場全景技術在虛擬實境中的應 用，例如：與Graphics-Based 3D Model的結合應用等。 The primary objective of this thesis is to explore the panoramic view generation. The user chooses a suitable viewpoint in real environment in advance, sets up the camera on a tripod , and then takes a number of (usually 24) images of the scene around the camera. Through some procedures of image processing such as the tilted image correction, image warping, image stitching, and image blending, we can warp these 24 images into a cylinder. We hope to provide an automatic procedure for the panoramic view generation.In case the user holds the camera with hands when taking the images, there might be some additional problems caused by camera pan, tilt andpitch. Under such circumstances, we can also correct these images by means of 2D projective transformation with the aid of the manually selected correspondence points. At the end, we can still make a panoramic view. We also do some virtual reality applications using the panoramic viewand combine the panoramic view with graphics- based 3D object models.