Improving or adjusting ORB performance

asked 2019-07-31 14:35:03 -0500

eric_engineer gravatar image

I'm using an ORB feature detector in Java with a camera feed. I don't seem to find as many points as SIFT (in python) did. And I notice that frame to frame the feature points it detects change, which makes my outline of the object a little "wobbly". So I'm wondering if there are any parameters I should be setting or tweaking to make it work better? Overall I think it is doing really well, I'm just trying to refine my results.

This is how I initialize it using opencv_346:

 detector = org.opencv.features2d.FeatureDetector.create(org.opencv.features2d.FeatureDetector.ORB);
    descriptor = org.opencv.features2d.DescriptorExtractor.create(DescriptorExtractor.ORB);
    matcher = DescriptorMatcher.create(DescriptorMatcher.BRUTEFORCE_HAMMING);

Then later I look through the matches and select those with a distance less than 40. A higher distance yields more points but more false points too.

Thanks for any advice!

edit retag flag offensive close merge delete


just curious - WHY areyou doing all of it ? what is the purpose ?

berak gravatar imageberak ( 2019-08-01 06:35:05 -0500 )edit