A recursive least squares solution for recovering robust planar homographies

Authors:
Saad Ali Imran;Nabil Aouf
Affiliations:
Department of Informatics and Systems Engineering, Sensors Group, Cranfield University, Shrivenham, UK;Department of Informatics and Systems Engineering, Sensors Group, Cranfield University, Shrivenham, UK
Venue:
TAROS'11 Proceedings of the 12th Annual conference on Towards autonomous robotic systems
Year:
2011

Citing 10
Cited 0

A survey of image registration techniques

ACM Computing Surveys (CSUR)
A robust technique for matching two uncalibrated images through the recovery of the unknown epipolar geometry

Artificial Intelligence - Special volume on computer vision
System identification (2nd ed.): theory for the user

System identification (2nd ed.): theory for the user
Evaluation of Interest Point Detectors

International Journal of Computer Vision - Special issue on a special section on visual surveillance
Fast panoramic image mosaicing using one-dimensional flow estimation

Real-Time Imaging
Multiple View Geometry in Computer Vision

Multiple View Geometry in Computer Vision
Distinctive Image Features from Scale-Invariant Keypoints

International Journal of Computer Vision
A Performance Evaluation of Local Descriptors

IEEE Transactions on Pattern Analysis and Machine Intelligence
Robust feature extraction and correspondence for UAV map building

MED '09 Proceedings of the 2009 17th Mediterranean Conference on Control and Automation
Video Image Mosaics in Real-Time Based on SIFT

PCSPA '10 Proceedings of the 2010 First International Conference on Pervasive Computing, Signal Processing and Applications

Quantified Score

Hi-index	0.00

Visualization

Abstract

Presented is a recursive least squares (RLS) solution for estimating planar homographies between overlapping images. The use of such a technique stems from its ability in dealing with corrupted and periodic measurements to provide the best solution. Furthermore, its capacity for providing reliable results for time varying parameter estimation also motivates its use in the context of real time cooperative image mosaicing where optimal transformation between mobile platforms is likely to change due to motion and varying ambient conditions and thus a way to tackle this problem real time is what is required. Additionally, and within the same context, a derived "match making" algorithm is introduced based on high curvature points (Harris points) and 3D intensity histograms which are in-turn matched using the L2 and L∞ and then compared to classical cross correlation(CC) techniques. Experimental results show that for synthetic data heavily corrupted by noise the RLS does a decent job of finding an improved homography, provided that the initial estimate is good. Results from real image data show similar results where the homography estimate is improved upon by periodic measurements. The match making algorithm proposed fairs well compared to intensity vector techniques, with the L∞ based method coming out on top.