Background
Visual landmarks
Landmark availability
Distinctive feature region extraction based on feature point grouping
Landmark candidates collection
Landmark selection procedure
SIFT feature matching
Landmark selection criteria
(a) Pairwise comparison of local feature regions
(b) Repeatability of local feature regions in input images
(c) Counting individual local feature correspondences in input images
(d) Using weight coefficient
Experiments
Settings
Quality landmark selection
-
Euclidean distance D to group two feature points ("Distinctive feature region extraction based on feature point grouping") was set to \(\sqrt{10}\).
-
Radius for investigating feature concentration ("SIFT feature matching") was set to \(d \le \sqrt{10}\)
-
Thresholds for the area of feature region \(S_s\) and \(S_l\) ("Landmark selection procedure") were set to 2500 and 40,000, respectively.