家用機器人視覺系統之相機校正

Abstract

相機校正是在電腦視覺中重建三維模型的一個很重要的步驟。我們必須要有正確的相機資訊，才能夠精確地重建出三維模型。相機校正一直是這一個領域很重要的研究主題，其方法大致可以分為照相測量術和自我校正兩類。在本篇論文中，將探討適用於家用機器人的相機校正方法。主要採用的方法是藉由在不同角度觀察一個平面的圖形，來推算相機的各個內部參數。在實驗中，虛擬影像和真實影像皆被使用，都能得到很好的結果，而且這個方法忍受雜訊的能力相當高。但因所需時間稍嫌緩慢，故考慮另一種利用Homography的方法快速地計算相機的焦距，並由結果的對應來獲得更精確的相機內部參數。

Camera calibration is a crucial step in the reconstruction of a 3D model and has been an important research topic in computer vision. We can classify calibration techniques roughly into two categories: photogrammetric calibration and self-calibration. In this paper, we will study different algorithms to calibrate a camera. The major method is based on images of a planar pattern obtained from different viewing angles, as proposed in [30]. Both synthetic data and real images have been tested and results with satisfactory accuracy have been obtained. The method deals with noise well but is time consuming. To improve the efficiency of the calibration, a second method which uses homography to quickly compute the focal length is adopted. Proper mapping between the results obtained by these two methods can then be used to derive the correct camera parameters.