基於機器人視覺之三維場景重建

Abstract

由二張以上同一場景的二維影像重建出三維的室內場景，這樣的技術在近幾年來已經被廣泛地應用到各個領域，包括機器人導航、電腦繪圖與虛擬實境等等。本篇論文即實作一套系統，基於家用機器人的應用需求，藉由一系列室內場景的二維影像，重建出三維景觀。本方法首先針對特徵點進行三維重建，由兩張以上的二維影像重建出多組的三維重建結果，再將多組重建結果整合為一，成為一個最佳的場景模型。接下來配合邊線擷取所得到邊線的資訊，進行線段的建立，將三維重建出之點連接成線。最後以半自動的方式，讓使用者透過圖形化的介面，選取出場景中主要的平面。我們希望藉由這套視覺系統，建立出場景的三維場景模型，以提供家用機器人或是其他相關系統之使用。

The techniques of 3D reconstruction from multiple 2D images of indoor scenes have been wildly adopted in many applications including robot navigation, computer graphics, and virtual reality. In this paper, we implement a robot vision system that can reconstruct a 3D scene from multiple views. First, the reconstruction of 3D feature points from 2D images is performed. Then, line segments connecting these 3D points are derived using the edge information obtained from the images. Finally, some major planes of the scene are identified semi-automatically with a graphical user interface. The 3D model of the scene established with the proposed vision system will hopefully provide useful information for home robot and other applications.