ORB vs FAST detectors

0
votes

I'm trying to detect Keypoints all over the image so I tried dividing it into cells and detect on each cell. However, I didn't get the same results using ORB detector as using FAST detector. For ORB, I get way lesser Keypoints as I increase the number of cells (smaller cells).

The picture below shows the results for dividing the image on 10 rows and 10 cols and max Keypoints 1000. The one on the left is the result for FAST ( 894 Keypoint) and on the right is the result for Orb detector (142 Keypoint).

Can someone explain to me why I get different results? Because I thought ORB is based on FAST features. And is there a way to get the same number of Keypoints as FAST while using ORB?

FAST Vs ORB detection

1
performanceopencvfeature-detectionorb
ORB will filter the keypoints returned from FAST so that the best keypoints are returned. You can try changing the fastThreshold to match the FAST detector and possibly the scoreType to ORB::FAST_SCORE. Can't say specifically what to set the fastThreshold to, you'd have to read the source code - EdChum
The problem is as shown in the picture enenthough I set the same fastThreshold for both detectors, I got way less keypoints with ORB detector. - user6099747
It'll still do additional filtering, the purpose of orb is different to fast I don't see why you would want orb to behave the same - EdChum
My purpose is to force orb to detect all over the image, so I divide it into cells and try detecting in each cell. The problem is that I'm not getting enough keypoints in each cell. - user6099747
I think you need to rethink your approach orb is going to detect features for a given aperture, so it finds the most distinctive features that are scale and rotation invariant if you just want to find features then you can just use fast as is, I unclear on what you really want to advice using this approach. There also an official opencv forum where you can post your problem answers.opencv.org/questions where you may get more luck - EdChum

1 Answers

0
votes

Eventhough ORB uses FAST keypoint detector it's not the same that we have to get the same number of keypoints while using the FAST and ORB.

ORB is just built on the FAST keypoint detector and the FAST detector is modified in ORB and it's not exactly the same(original one). In the official paper of ORB it states the additional contributions to the FAST detector in ORB and kindly have a look into it. 

"FAST does not produce a measure of cornerness, and we have found that it has large 
responses along edges. We employ a Harris corner measure [11] to order the FAST keypoints. 
For a target number N of keypoints, we first set the threshold low enough to get 
more than N keypoints, then order them according to the Harris measure,
and pick the top N points. "

This might be one of the reasons why it is giving you less number of keypoints. What I can suggest you from my end is just minimise the threshold value and so you can get more number of keypoints.