從物體影像產生線光源的影子

Abstract

本篇論文提出了一個嶄新的方法使用2D的資訊來產生物體的影子。在以往，影子的做法通常都是以3D幾何為基礎。在此論文介紹的方法中，要產生一個3D物體正確的影子，基本上只需要兩張物體不同視點的影像。對於物體的每個視點而言，當一個點光源落於上述兩個視點連線上時，物體的影子都可以被產生出來。而這個方法使我們可以利用取樣的方式，以一連串的點光源來模擬一條線光源的效果。這個方法假設在物體投射影子的平面上，至少有五個且任意三個不共線的參考點可以在影像上被定位出來。接著，根據參考點的位置，利用交比的透視投影不變性以及一些透視投影的幾何特性，即可找出物體在投射平面上的影子區域。這個方法可以很容易的被推廣至更多的問題上，例如多個物體以及多個投射平面等情形。

This paper presents a novel approach to generating object shadows using two dimensional（2D）information. Previous approaches of shadow generation are usually based on 3D geometry. In this paper, we presents a way of deriving accurate shadows of a 3D object which requires only stereo images of the object taken from two different viewpoints. For each view of the object, the object shadows can be generated for point light sources located along the line connecting the two view points. The proposed approach can thus use a series of point light sources through sampling to simulate a linear light source. The approach assumes that at least five reference points on the base plane, where the object shadow is to be cast upon, can be identified in the stereo images. The shadow generation is accomplished by identifying the portion of the base plane which is in shadow in an image with respect to the reference points using the view-invariant cross ratio and some geometric properties associated with perspective projection. The approach can be extended easily to solve more general problems which involve multiple objects as well as multiple planar surfaces on which object shadows are to be cast.