L4: VSLAM 2

Question 1

Explain Graph-SLAM approach

Question 2

What is the purpose of front-end and back-end in SLAM?

Answer 2

Front End → VO

Back End → BA
Interplay of front-end and back-end!

Purpose: VO is computationally light, we can do it online. BA is not (except if windowed) so it can be done when there is time or if necesarry.

Question 3

Feature detection

Answer 3

Examples:
- FAST

Question 4

Feature description

Answer 4

Examples:
- SIFT
- ORB
- BRIEF

Question 5

SIFT

Question 6

FAST

Question 7

ORB

Question 8

BRIEF

Question 9

Feature matching

Answer 9

Find correspondences between images by solving the correspondence problem. By comparing the features with their descriptors.
Examples:
- BF-Matcher: Check the whole data, find the smallest distance, gives the best result. Very slow for big data.
- FLANN based matcher: Fast, gives acceptable results, great with big data. Use euclidean (kd-tree) or hamming distance between descriptors. Not as robust as BF-Matchers.