52 / 56
(2)
Performance/Speed Tradeoff:
(3)
Consistency:
Sum-up:
GMS achieved real-time matching of features in challenging scenarios, not yet
successfully dealt with before. It shows high performance, with much better
speeds than comparably performing feature extraction methods.
Research
Compared to 9 other methods,
GMS shows performance near
that of the best, and speed
consistently among the
quickest. None of the other top-
performing methods show
speed near that of GMS. This
can be particularly useful for
real-time applications, where
speed may be the more critical
consideration in judging
algorithms.
The figure below illustrates performance variation across different TUM
scenes. The red mark is the median, the blue box covers the 25th to 75th
percentiles. Whiskers show performance beyond the 25th and 75th
percentiles. GMS (the fifth from the left, labelled in red) is the most consistent
fast algorithm, its consistency comparable to that of much slower ones.
52
Computer Vision NewsResearch