How can I use features SURF/ORB/... to detect object with brightness-invariant?

3
votes

How can I use one of feature types to detect object - illumination / brightness invariant?

Interested to use features that resistant to:

  • different lighting
  • half of the object in the shadow
  • glare/reflections

Does it make sense to use a HUE (1st component of HSV-color-space), or the average value between the HUE and brightness?

And what is the best feature SIFT/SURF, ORB, BRISK/FREAK, KAZE/AKAZE for brightness-invariant detection?

1
opencvcomputer-visionfeature-detectionfeature-extractionsurf

1 Answers

4
votes

Sorry for the late response but this answer might be beneficial to someone else.

This is quite a problematic area that I am also facing. Unfortunately I don't know of any feature detector-descriptor combination that is illumination invariant. So some suggestions that you might want to consider involve the following:

  1. You can pre-process the images using the Wallis filter so that both images get to have a balanced brightness level throughout the images.

  2. You can also normalize the descriptor values of the detected features since the descriptors use image intensity values to construct the descriptors. So if the images have different illumination conditions, then corresponding features will also have different descriptors.