Feature Matching

Feature matching solves the data association problem in SLAM: determine the current view correspondence between the landmarks (feature points) and the landmarks (feature points) seen before.

1. Brute-Force Matcher

Step 1: Measure the distance between each feature point $x_t^m$ and all $x^n_{t+1}$ descriptors, and then

Step 2: Sort by degree of similarity of two features.

​ For descriptors of floating-point type, use Euclidean distance to measure

​ For binary descriptors (such as BRIEF), use Hamming distance : the number of different digits.

Step 3: Take the closest one as the matching point.

2. Fast Approximate Nearest Neighbor (FLANN)

When the number of feature points is very large, we can use Fast Approximate Nearest Neighbor.