多物體格子編碼影像之物面自動分割

Abstract

在計算機視覺領域中，由一張二維 (2-D) 影像來作三維 (3-D) 物體之構 建及識別是一個重要的研究課題。而利用格子編碼技巧，所發展出來的主 動感測式計算機視覺方法，其基本步驟是先以結構光源將網狀格子投射在 多個三維物體表面上，其次經相機取得影像後，再針對此具有網格線之影 像，利用一定程序求得網格線及交叉點位置，並進而推導出三維物體的位 置及相關參數。本篇論文提出一套方法，以處理多個物體且允許部份相互 遮蔽的格子編碼影像，希望自動產生網格線及網格交叉點位置，以作為結 構光源技巧之前置處理。對於處理多個三維物體的格子編碼影像，除了追 蹤出網格線外，我們尚提出一套簡單的方法，以判斷網格交叉點間所形成 的線段之不同性質，如長度、斜率等，去除多餘交叉點並進一步作物體分 割，使不同物體面得以分開。同時由所得到之物體各表面之網格線及交叉 點，作為未來對部分物體之構建與辨認之用。 In the field of computer vision, the reconstruction and recognition of a 3-D object from a 2-D image are important research topics. The active sensing grid-coded techniques consists of four basic steps. The first step is to use the structured light to project grid pattern onto the surfaces of multiple 3-D objects. The second step is to take the images with a CCD camera. The third step is to find the positions of grid lines and their junction points. The final step is to derive the location and related parameters of the 3-D objects. In this thesis, we propose a method to extract multiple object surfaces from a grid-coded image that allows object mutual occlusion. Also, the positions of grid lines and their junction points can be found automatically. These grid information will be used in the structured lighting technique. In segmenting the grid-coded images, we present a method to use the length and slope of each line segment. By using these features of each line segment, we can decompose the grid lines into different object surfaces. After we obtain the grid information of each object surface, we shall develope algorithms to reconstruct and recognize the objects.