Automatic Matching of Close-Range Video Images Using Parameter Space Clustering

Abstract

Image matching, which amounts to the automatic establishment of the correspondences between two images or more, is a fundamental problem in digital photogrammetry. It has a large number of applications such as image mosaicing and 3D surface reconstruction from images. The contributions of this paper are two folds. First, it presents a robust strategy for point features selection. Second, it presents a novel method for automatic point features matching for the images that were extracted from a moving video camera. The proposed matching methodology uses point features as matching entities and parameter space clustering as a matching method. The basic idea underpinning the parameter space clustering methodology is to pair each data element belonging to two overlapping images, with all other data in each image, through a mathematical transformation. The results of pairing are encoded and exploited in histogram-like arrays as clusters of votes in the parameter space defined by the transformation function. Due to the nature of video images the mathematical transformation that defines the parametric relationship between the two images is approximated by a 2D translation. As a consequence of this approximation, the matching problem is approached as an inexact-matching. The maximum consistent subset of votes in the parameter space is exploited to reveal the underlying correspondences between the two images. Successful and promising experimental results of matching video images are reported in this paper.