Robust Method of Recovering Epipolar Geometry Using Messy Genetic Algorithm

• Venue: CRV '04 Proceedings of the 1st Canadian Conference on Computer and Robot Vision
• Year: 2004

This paper addresses the problem of robustly estimatingthe epipolar geometry by employing a new techniquebased on messy genetic algorithms, which uses each geneto stand for a pair of correspondences, and takes everychromosome as a minimum subset for epipolar geometryestimation. The method would eventually converge to anearly optimal solution and is relatively unaffected byoutliers. Experiments with both synthetic data and realimages show that our method is more robust and accuratethan other typical methods because it can efficiently detectand delete the bad corresponding points, which includeboth bad locations and false matches.