Feature Matching

Fast Library for Approximate Nearest Neighbours (FLANN)

I heard about this from PCL. Using it in my Visual SLAM project.

Resources

• https://docs.opencv.org/4.x/d5/d6f/tutorial_feature_flann_matcher.html

FLANN is a faster matcher than Brute-Force Matcher.

How much faster is FLANN than BF-Matcher?

Just search for time complexities of kd-Tree.

How it works? With the brute-force matcher, it runs in $O(n^2)$ since it does a pairwise comparison. With FLANN, we essentially preprocess the data, organizing it using a kd-Tree / K-Means tree for higher dimension.

Because we make use of this smart data structure, we can search for nearest neighbors much faster.