1

I want to use hog detector in order to know if a ROI is a person or not, I already coded a filtering technique allowing me to select ROIs that can contain a person.

for instance this is what I'm doing: 1.saving my rois as rectangles 2.for each roi: -cropping the original image, rescaling it and then applying hog.detect() on this image instead of the original one.

this is working but it doesn't seem to be the right way to do what I want. I found out lately that there is the hog.detectROI() function, and I'm wondering if I can use it and give it my image and my ROIs as input only once, will it be able to tell me wich ROIs are people? I would be grateful if someone can give me an example to know how to use this function, especially what should I give as locations when my ROIs are rectangles?

자스민
  • 21
  • 4
  • Usually, one first selects a problem and then the methods to solve it. Choosing a "hog detector" first and the "detect person" later is reversing the steps. That's not to say it's impossible, just that you have an unusual approach to problem solving. – MSalters May 02 '16 at 10:22
  • 1
    It's indeed clearer. The method seems to take points, not rectangles. Have you tried passing the center of your ROI's? – MSalters May 02 '16 at 14:56
  • That's right , I thought about the center or the top-left corner, but my ROI's are not the same size. and I'm not sure if the detectROI () has the output that I'm looking for. In the opencv documentation its description is: "evaluate specified ROI and return confidence value for each location ", what does this mean? – 자스민 May 02 '16 at 21:02

0 Answers0