室內場景之特徵點擷取與追蹤

Abstract

由一對立體影像或是多張連續的影像中找出特徵點的對應是電腦視覺中一個經典且困難的問題。本篇論文提出了一個演算法來解決特徵點對應的問題，並設法減少計算量以期能夠達到機器人視覺中建構室內場景所需的即時性。在我們所提出的方法中，首先使用Harris演算法來擷取影像中的角點作為特徵點，接著分析在Harris計算過程中所得到的影像梯度，將影像梯度方向做分類並量化，再分別以不同的顏色來代表所量化的區域，最後所得到的這些彩色碼即可成為此角點所具有的特徵。對每一個特徵點找出其所對應的彩色碼之後，後續的比對工作便可經由彩色碼的比較來達成：若兩點的彩色碼越相似則表示此兩個角點非常有可能是互相對應的兩點。透過距離限制和雙向對應一致性的要求，以及與相鄰點關係的比較，並計算出一評估對應關係的分數，最佳的對應點即可被選擇出來。實驗結果顯示我們所提出的特徵點對應演算法是具有一定的正確性和效率性的。

Establishing feature point correspondences from a pair of stereo or a long sequence of images is a common and critical problem in computer vision. We propose an algorithm to solve this problem and hope to achieve real-time performance by reducing the amount of calculation of the vision system. According to the proposed algorithm, we first extract corner points from images as feature points by the Harris corner detector. The image gradient obtained from Harris corner detector is then classified it into 9 regions according their gradient directions and magnitude. The quantized gradient directions are represented as different colors for each feature point to form a color code. Finally, the point correspondences are obtained by comparing the color codes as well as the spatial relationships between neighboring feature points. Experiments show that the proposed algorithm is efficient and very robust for the matching of feature point.