Homography Ranking Based on Multiple Groups of Point Correspondences.

Milan Ondrašovič,Peter Tarábek

doi:10.3390/s21175752

Abstract

Homography mapping is often exploited to remove perspective distortion in images and can be estimated using point correspondences of a known object (marker). We focus on scenarios with multiple markers placed on the same plane if their relative positions in the world are unknown, causing an indeterminate point correspondence. Existing approaches may only estimate an isolated homography for each marker and cannot determine which homography achieves the best reprojection over the entire image. We thus propose a method to rank isolated homographies obtained from multiple distinct markers to select the best homography. This method extends existing approaches in the post-processing stage, provided that the point correspondences are available and that the markers differ only by similarity transformation after rectification. We demonstrate the robustness of our method using a synthetic dataset and show an approximately 60% relative improvement over the random selection strategy based on the homography estimation from the OpenCV library.

Highlights

Homography is a perspective projection of a plane from one camera view into a different camera view
Homography is commonly used for the rectification of text document images by generating a fronto-parallel view [5,6], image stitching [7,8], video stabilization [9], extracting metric information from 2D images [10], and pose estimation [11] and for various traffic-related applications, e.g., ground-plane detection [12] and bird’s-eye view projection [13]
We evaluated our homography ranking in terms of reprojection error improvements against the existing approaches based on the isolated homography estimation represented by OpenCV [19] implementation

Summary

Introduction

Homography is a perspective projection of a plane from one camera view into a different camera view. The perspective projection maps points from a 3D world onto a 2D image plane along lines that emanate from a single point [1,2]. Homography estimation is essential for image registration, i.e., a process of image matching and transformation of two or more different images [14] It can be addressed either on the pixel or feature levels. A single marker is identified in the image by multiple independent keypoints that have a direct correspondence to its real shape, making a group of point correspondences. These correspondences are often noisy and they can introduce errors in the homography estimation.

Objectives

Methods

Results

Conclusion