Skip to content

Simple library for sparse feature matching and homography estimation.

Notifications You must be signed in to change notification settings

bjh83/Feature-Matching

Repository files navigation

Feature Matching And Homography Estimation

This is a purely educational project attempting to create a simple library for sparse feature matching and homography estimation. The short term goal is to learn more about feature matching. The long term goal is to create a suite of tools that allow me to do experiments with projective and camera based multiple view geometry.

Feature Detection

I use a simple Forster-Harris feature detector (corner detector as some call it) to measure the 'goodness' of a point as the center of a feature. I then reduce the 'corner score map' by doing a non-adaptive non-maximal suppression sub-sampling. Since the long term goal is to use point correspondences to create a homography describing the relationship between the given images, I do not need a terribly large (or consistent) number of point correspondences; really, the things that matter are that the point correspondences are high quality (accurate and precise), that they cover a wide range of the image, and that they are not collinear.

Feature Description

I am currently in the process of implementing the MOPS feature description/matching algorithm as described in Brown, Szeliski, and Winder's "Multi-Image Matching using Multi-Scale Oriented Patches."

Homography Computation

The plan here is to use the normalized Direct Linear Transformation (DLT) using the point correspondences computed above. I am pretty new to this stuff and I have no idea how well this will work aside from some reading, so we will see how this goes and maybe try an iteratively based refinement of some sort.

About

Simple library for sparse feature matching and homography estimation.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages