Authors: Chunmei Duan, Xiangxu Meng, Chenglei Yang
Addresses: School of Management and Economics, Shandong Normal University, Jinan, China. ' School of Computer Science and Technology, Shandong University, Jinan, China. ' School of Computer Science and Technology, Shandong University, Jinan, China
Abstract: In this paper we propose a method to improve feature point matching results for structure from motion problems, which includes techniques for promoting the correspondence quality using epipolar geometry, fast decision from relative positions of matches and local geometry constraints. We start with feature points extracted from each image and initial set of matches obtained by cross correlation and relaxation principle, then several strategies exploited to detect outliers. Our method is performed before projective reconstruction phase and only two views of scene are needed. Experiments have been carried out and the results demonstrate the efficiency of proposed strategies.
Keywords: feature point matching; fundamental matrix; epipolar geometry; 3D reconstruction; image matching; 3D modelling; motion; computer vision.
International Journal of Computer Applications in Technology, 2010 Vol.38 No.1/2/3, pp.101 - 105
Received: 04 Aug 2008
Accepted: 05 Jan 2009
Published online: 16 Jul 2010 *