Year: 2007
Type: other
Abstract: At the same time, however, the random sampling stage is very time-consuming.It may take many random samples before a set of inliers is found.For this reason, two methods (Chapter 3 and 4) have been developed to guide the sampling process.Instead of a purely random search, the search will be directed by using non-uniform sampling probabilities for the correspondences.Another reason for the lengthy sampling stage, is the size of the sampled point set.The larger the point set, the longer it takes before an inlier is sampled for each of the points simultaneously.Therefore, a method (Chapter 5) has been developed that samples smaller-than-minimal point sets for estimating the fundamental matrix.Although the proposed methods are directed towards fundamental matrix estimation, some of them have a larger extent.In particular, the methods in Chapter 4 and 5 are also applicable to hyperplane estimation.Line and plane fitting in 2D and 3D, respectively, will serve as examples in these chapters.The methods described in Chapter 3 and 4 assume the presence of a single (dominant) epipolar relation.This holds when there is no independent movement of objects between two recordings.If objects do move independently, then multiple
Source: TNO Repository
Author: Richard J. M. den Hollander
Language: English
Topic: Advanced Vision and Imaging
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2139770562