Research On Image Matching Based On Epipolar Geometry

Epipolar geometry is the same scene that the geometric relationship between the two images, it is independent of scene structure and only depends on the camera parameters, it is the nature of the inherent projective between two images. Research on epipolar geometry, could be widely used in image matching,3D reconstruction and other fields.Based on the epipolar geometry of the image matching,the fundamental aim is to recover the epipolar geometry. Epipolar geometry can be described in a 3×3 fundamental matrix, the recovery of the epipolar geometry can change to estimating the fundamental matrix. However, error data and estimates the time required for estimating the fundamental matrix has been troubled by the noise interference.Based on this, in the extracting feature stage, original feature points detection algorithm for the poor robustness characteristic. Improve the Harris feature point detection algorithm: First use of the image histogram equalization to increase the image contrastion, rich local image feature information;Using median filter and Gaussian filter to enhance the robustness of the algorithm;Adaptive method to determine the threshold, removing a part of pseudo-feature points which characteristic information are smaller than the threshold. Pave the way for the subsequent algorithm.In the feature point pre-match stage, using the normalized cross-correlation matching algorithms, on feature point set for pre-match obtain pre-matched set of points.In the fundamental matrix estimation stage, Adopt good robustnessand the most widely used RANSAC((Random Sample Consensus) algorithm to remove error data. Through research and analysis RANSAC algorithm, For its number of inaccurate sampling, sampling time is too long, the rough threshold detection of defects, Improvements to enhance the efficiency and robustness of the algorithm. Pre-processing algorithm using the improved set of matching points, estimate the fundamental matrix. Finally, by analysis and verification, experimental results show that the improved algorithm efficiency, robustness.