環物影片產生及應用技術之研究

Abstract

由於環物影片(Object Movies)在建置上相當簡單以及能產生相片品質的成像結果，因此環物影片目前已是一種非常通用的方法來表現可互動的三維物體。雖然環物影片已成功地被應用在不同的領域，但關於產生高品質環物影片的技術以及更具有開創性應用的技術都有待研究開發。 本論文提出一些方法來產生高品質的環物影片以及展示一些利用環物影片來表現三維物體的應用。首先，我們提出一個環物影片拍攝架的挍正方法並提供一個視覺化的界面幫助使用者根據計算結果來調整拍攝架。實驗顯示利用此校正方法，只需取得12張校正物的影像便可以得到任一視角的精確相機參數。接著，我們提出環物影片背景去除的方法。該方法最主要的好處是使用者的介入非常地少。更確切地說使用者只需從自動去背景後的影像中選擇足夠好的結果，該方法便能自動地將這些正確的資訊傳遞到其它的影像並修正去背結果。在傳遞的過程中，該方法利用選擇的影像重建該三維物體的三維幾何形狀，再投影到其它的影像上產生物體的剪影以協助自動去背方法產生更好的去背結果。實驗顯示，透體這剪影資訊可以明顯地降低去背結果的誤差並能有效地抵抗雜訊的影響。另外，一個新的三維重建方法被提出來從環物影片之中重建出該物體的三維資訊。相較之前的方法，該方法所產生的三維模型可以保留住更多的細節部份而且其剪影會和原始輸入的剪影影像相吻合。最後，我們延續之前環場環物整合技術的研究，提出可將立體環物整合到立體環場的方法。使用者可以透過所提供的系統可以整合立體環場環物建置出一個非常逼真的互動環境，並可直接瀏覽該虛擬環境以及觀看立體環物。經由這樣的互動環境使用者可以透過立體視覺體驗到更真實的3D感受。

Object movie (OM) is a conventional approach for modeling and rendering interactive 3D objects because of its simplicity in production and its photorealistic presentation of objects. Although OMs have been successfully adopted in many applications, the techniques for production and application of OMs must still be enhanced if high-quality and efficient OMs are desired. This work proposes some methods for generating high quality OMs, and demonstrates some applications using generated OMs to present the 3D objects. First, a method for calibrating the motorized object rig is presented, and a visual tool is introduced to adjust the axes of the motorized object rig. The distances among all the three axes of the motorized object rig can be minimized after adjustment, and more reliable camera parameters can be obtained after the calibration process. Experimental results indicate that highly accurate parameters can be obtained from only 12 images. Second, an image segmentation method is proposed to remove the backgrounds of OMs. The major advantage of the proposed method is it can propagate the successful segmentation results from some selected images to the whole OM. The new OM segmentation method extracts a 2D shape from the reconstructed 3D model and uses the 2D shape to remove the background from the foreground object. This work demonstrates that the proposed method can significantly improve OM segmentation. Third, a novel approach is proposed to reconstruct high quality 3D models from OMs. The silhouettes and detail features of reconstructed 3D model are successfully preserved. Finally, previous work on augmented panoramas is extended to augmented stereo panoramas. This work develops an interactive system that allows the user to integrate stereo OMs into a stereo panorama, and interactively browses the augmented stereo panorama. To generate stereo OMs, the 3D models reconstructed from monocular OMs are rendered. The proposed method takes less than half of the processing time, including acquisition and segmentation, than traditional approaches, which take two separate sets of OMs. The proposed interactive system provides the users with two approaches to determine the reference frames where the object is inserted in a stereo panorama. The left view and the right view are rendered separately after determining the reference frames. For each view, the background layer is first rendered, followed by the shadow and the object layers. A user can directly rotate and translate the stereo object movie of interest by browsing the augmented panorama. The augmented stereo panoramas provide users with more persuasive interaction with better depth perception.