利用變焦影像產生高動態深度範圍之深度圖

Abstract

隨著立體形象化技術的進步，在日常生活中，不同的娛樂方式如雨後春筍般蓬勃發展。人們可以看立體電影或與遊戲中的物件做三維互動，這的確是近年來全新的體驗。科學家與研究者也一直希望能發展一套系統可以提供人們有更真實的感受，然而無可避免地，我們渴望能夠顯示更豐富的內容，只是間接獲取深度資訊的技術時常受限於影像的內容，對消費者來說，利用相機陣列拍照也是不切實際的。所以我們選擇了實用的單相機系統，並且也利用透鏡陣列來擷取具有視差的基本影像。除此之外，我們之所以利用深度圖來表示我們的三維資訊是因為深度圖可以被應用於一般二維多工式的三維顯示器的立體影像組合成。但能夠分析視差的前提是基本影像需要是全聚焦的。對一般攝影來說，調整光圈來延展景深是最直接的方式，然而光圈太小時，當場景是昏暗的以及有移動的物體時，會碰到曝光時間的兩難。所以我們承繼了高動態範圍影像的概念，發展了一套高動態深度範圍的系統來產生一張具有廣闊工作範圍的深度圖。我們所提出的高動態深度範圍系統不僅提供了較我們相機裡最小光圈(150cm)更廣的深度範圍(165cm)，並且也縮短了至少21倍的曝光時間。而這樣的特性可以使得我們可以擷取快速移動的物體，卻不受一般相機大光圈的淺景深所限制。

With the improvement of 3D visualization, many recreational activities spring up in our daily life. People can watch 3D movies or interact with 3D video games. It is indeed a brand-new experience around these years. A lot of scientists and researchers keep working on developing a system that can provide humans with more realistic perception. To accomplish this great task, longing for more abundant 3D content is inevitable. However, indirect rendering of depth information is often limited by the image categories, and it is impractical for customers to use camera array to create their own 3D content. As a result, single camera system is chosen for its practicability and we also utilize lens array to generate disparity in the elemental images. Moreover, we represent our 3D information by a depth map because it can be used to synthesize stereo image pairs for conventional multiplexed-2D 3D displays. Nonetheless, the prerequisite of stereo matching is all-in-focus elemental image. Increasing f-number to elongate depth of field is the most common way in photography, but this strategy confronts the dilemma of exposure time when the scene is dim and contains moving objects. Hence, we propose a high dynamic depth range (HDDR) system to render a depth map with wide working range, and this idea stems from the concept of high dynamic range (HDR) images. Our proposed HDDR system not only provides wider working range (165 cm) than that of the largest f-number in our camera (150 cm), but also minimizes the exposure time by at least 21 times. And this characteristic will make it possible to capturing the object with quick movement but without limitation of the shallow depth of field of small f-number.