Fast SIFT Feature Matching Algorithm Based on Geometric Transformation

Abstract

Structure from Motion (SfM) is a series of methods for reconstructing scene structure (i.e.three-dimensional space points) and camera motion (i.e.camera pose) from an image set. In this paper, aimed at the low accuracy of the global reconstruction, the robustness to external points and the time-consuming incremental reconstruction, a fast, closed-loop and high precision reconstruction method is proposed. The method is based on the SIFT matching algorithm GeoMatch, which is constrained by geometric structure of the scene and by numerical and statistical characteristics of feature invariant scale transformation. Experiments show that GeoMatch outperforms both traditional tree-based and hash-based matching methods in terms of time and accuracy.