How to improve object detection using Haar Cascade Opencv C++?

asked 2017-11-15 02:09:24 -0500

vbc gravatar image

updated 2017-11-15 02:10:55 -0500

I had using haar cascade to detect object( plate of car). But confidence not high. Example if when we have picture contain plate but algorithm can't detect plate. How do i can retrain to detect the missing plate? Anyone know the way to improve it?

edit retag flag offensive close merge delete


you can only re-train cascades, if you have the original data.

if you find out, that there's too much variation in your "real life" data, it's probably an argument against using cascades for this at all.

berak gravatar imageberak ( 2017-11-15 02:38:24 -0500 )edit

@berak: your mean is if i retrain it, i can re-train failure data but i still can't detect plate on this data? What solution to detect plate on this case?

vbc gravatar imagevbc ( 2017-11-15 03:37:12 -0500 )edit
  1. all you can try is: train the cascade from scratch, using more & better data
  2. i have not seen anyone being really successful with the general idea, yet. (maybe it's the wrong tool ?)
berak gravatar imageberak ( 2017-11-15 03:45:17 -0500 )edit
sturkmen gravatar imagesturkmen ( 2017-11-15 12:11:35 -0500 )edit