leesjpower's profile - activity

Hi all,

I'm a newbie of opencv and recently I've done a project using opencv_traincascade with LBP feature.

I have around 2000 pos samples and also around 2000 neg samples.

And with these, I tried to make 18~20 stages of xml files with different options.

The result of one of these was not too bad but I want to increase the detection rate and decrease miss rate more than now.

So I think that, if I choose correct pos samples, my code can detect more correctly and more frequently. Which one is about good samples?

1. high resolution about 800x600, pos object with large random background(small ROI)
2. high resolution about 800x600, pos object with small random background(big ROI, entire picture)
3. high resolution about 800x600, pos object without background(removed background, no white, just removed)
4. low resolution about 48x48, pos object with small background(big ROI, entire picture)
5. low resolution about 48x48, pos object without background(removed background again)
6. other ideas

(sorry about my Eng if I have mistakes or you cannot understand my question)