利用以消失點為基礎之新模型從單一視角進行立體視訊合成

Abstract

隨著現今立體顯示技術(多視角影像擷取器材以及相關的壓縮編碼演算法)的發展，它所帶來的立體視覺體驗，已經令使大家逐漸地產生興趣。然而為了產生具有深度的感受，無論是拍攝或者以單一視角/多視角合成的方法，產生雙眼視角的兩隻不同影片是必要的。在這篇論文中，我們提出一個從單一視角產生立體影片的方法。這個方法先將背景和物體分割後，然後再進行座標軸轉換來合成立體影片。我們提出的演算法是利用單一視角影片中的消失點和消失線所產生的立體空間模型，進行影像座標和世界座標之間的轉換，經由這個轉換後，再模擬出另一視角攝影機的位置，進行反轉換產生立體影片。最後，我們除了會放上產生的立體影像來展示立體效果外，還會將合成的影像和真實影像以PSNR進行比較，以此來客觀地表示以此方法產生影片的相似程度。

With the recent progress of stereoscopic display technology, multiview capture devices, and the corresponding coding algorithms, stereoscopic viewing experience has been introduced to the public with growing interest. To create the depth perception, two different video sequences in binocular vision are captured or synthesized from the monocular/multiview video sequences. In this paper, we produce the stereoscopic videos out of the monocular sequences by means of the background/object separation and the coordinates transformation. The proposed algorithms transform the image coordinate into the world coordinate based on the vanishing model constructed from vanishing point and vanishing lines of the monocular video. At last, we would exhibit the experimental results to show the stereoscopic effect and compare the synthesized views with raw data in PSNR to represent the similarity from the proposed algorithm.