特徵點的擷取與對應

Abstract

在兩張或是一連串連續的影像上擷取特徵點及找尋其對應關係，已成為三維重建或相機校正等應用所不可或缺的一項技術。本篇論文設計出一套簡單自動擷取特徵點並找尋對應點的系統，其目的在對於兩張或是一連串連續的影像中找到至少八組以上的正確對應點。本篇論文所採取的方法不需要事先了解拍攝場景或是拍攝的機器特性，但有一個前提是連續兩張影像的相似程度要高，或是拍攝此兩張影像的時間必須很接近，因為若兩張影像的差異太大，則要找尋正確的對應點是相當困難且費時的。本篇論文利用邊協助做特徵點的群組化及驗證對應點的正確性，實驗結果顯示如此作法在正確性與計算效率兩方面均有不錯的結果。

Many computer vision tasks, e.g., camera calibration and 3-D reconstruction, rely on feature points extraction and matching. In this thesis, we design a system that can automatically extract feature points and find the matching pairs between two images or in an image sequence. The goal is to provide at least 8 matching pairs for the computation required in camera calibration and 3-D reconstruction. The system doesn’t need any prior information about the scene or the camera. The only constraint is that any two successive images should be fairly similar. Otherwise, feature points matching will be very difficult and time-consuming. The system simulation results shows that the correctness and the efficiency of the feature point extraction and matching are improved by such an approach which uses the edge information to assist the matching process.