Adding hog feature to KCF tracker [closed]

edit

you've seen this tutorial ?

(please try and report back !)

but oh, wait. it's expecting a feature in the whole size of the image, not an "embedding", like HOG.

(but you could still try adding "gradient / magnitude" features (instead of "binning it on patches", like HOG does)

I'm using Python, maybe it's impossible to declare parameters to tracker because it says 'unexpected arguments'. Do you know any Python tut for this? tracker = cv2.TrackerKCF_create(cv2.TrackerKCF_GRAY | cv2.TrackerKCF_CN, 0, True,2 )

the last 2 letters iin KCF stand for "color features" (it's like blowing up 3 RGB colors into 9 channels, using a lookup table, and if you do that on grayscale, you're actually defeating that

being unable to enforce "grayscale images" from python is might be a minor problem here. (rather look up the whitepaper on this)

(and also, while you can't setcv2.TrackerKCF_GRAY from python (if i recall this corectly), feeding a grayscale image* into the algorithm will achieve the same.

if you want my 2ct.:

write your own tracker (there's a VERY successful MOSSE example here.

Thanks for answers, I'll look at example and try to make my tracker, otherwise it doesn't work :)