以軌跡為基礎之立體視訊穩定演算法

Abstract

鑒於立體電影「阿凡達」的成功，以及對應螢幕和相機的問世，揭 櫫了立體多媒體世代的來臨。然而相較於立體硬體設備的快速發展， 相關的的軟體技術卻鮮少被提及。這篇論文由立體視訊穩定著眼，將 原先應用於平面多媒體的技術，擴展至立體多媒體領域。今日，立體 映像的基本原理為，讓觀眾於同一時間，左右兩眼分別看到不同的影 像，使其於螢幕前或後一定範圍聚焦而達到立體的效果。亦即，若要 處理立體多媒體影像，必須同時對兩個影像做處理。然而這兩張影像 之間須符合一些針對立體映像的特殊條件，包括顯示比例設計以及空 間座標的定位，否則使用者將難以對畫面聚焦，而無法產生立體感。 更甚者，將會產生噁心、暈眩等副作用。若不考慮上述條件，直接將 用於平面多媒體的處理技術，套用於立體多媒體上，將會破壞立體映 像的型塑要件。這篇論文將對立體映像的條件定性，配合以軌跡為基 礎的視訊穩定演算法，提出更貼近人體視覺的立體視訊穩定演算法。 使用者測試的結果表示，相較於直接套用平面多媒體技術，我們的方 法更能讓使用者舒適地享受立體視訊。

The success of the 3D film Avatar and the emergence of stereoscopic displays and binocular cameras signify the advent of the 3D multimedia era. However, despite fast development of stereoscopic hardware, less progress has been made towards the software side. This paper represents a step towards extending conventional 2D media operations to 3D media editing by addressing the problem of stereoscopic video stabilization. Although a stereoscopic video is no more than a pair of videos respectively corresponding to two eyes, there exist stereoscopic relations between two views. Without respecting/ maintaining to these relations, the resulting imagery could become blur or even not stereoscopic at all. Thus, naive applications of conventional 2D image/ video processing methods on imagery of both views independently could destroy the correlation between views required for 3D fusion. This paper incorporates stereoscopic constraints into trajectory-based video stabilization methods to maintain viewing comfort and pleasure to facilitate processing of stereoscopic videos. User studies show that our method provide more comfortable and enjoyable 3D viewing experiences than naive extensions of conventional stabilization methods could deliver.