0

I used Hough transform in the same image to obtain more lines of the same images. Now I want to match these lines to have, at the end, a few lines that identify my target.

Jonas Stein
  • 6,826
  • 7
  • 40
  • 72
Leonard993
  • 1
  • 1
  • Welcome to SO. Could you please edit your question and add extra information, example images, a bit of math/code? As it stands what you want is very, very unclear... – Ash Oct 06 '18 at 09:55
  • My problem is to extract feature from image like this: https://www.dropbox.com/s/884ibbrh23z7z7n/Out.png?dl=0. I'm using hough function but it's hard for me to chose a correct setting to obtain a best fit of image. So I divided the problem in three step. – Leonard993 Oct 06 '18 at 10:19
  • The first is to find lines through hough function with three different settings, the second step is to merge all these lines like the following image: https://www.dropbox.com/s/ir2rzxqp3aonqdy/figure.png?dl=0. The third step is to obtain the best configuration of lines to obtain the target. The problem is that I have many overlapping lines and I don't know what can I do. I hope that now is clear. Thanks. – Leonard993 Oct 06 '18 at 10:19
  • It's a bit better thanks. So the target is the planar looking object? Looking at your images, I think that you should question the entire pipeline, not just the third step. It seems to me that it would be better if you tried focusing on the geometric/photometric properties that are unique to the target instead of lines, that can be found everywhere else. The target is planar, why not use some monocular plane detection? Also, that kind of segmentation is something that a neural network would be good at. Is that image part of a video stream? If yes then there are many possibilities... – Ash Oct 06 '18 at 10:54
  • Also, just as a side note, please refrain from adding dropbox links (they attract downvotes very fast, understandably), and please embed images/extra info in your question itself instead of comments. – Ash Oct 06 '18 at 10:58
  • You're right. I can't use the geometric/photometric properties because I have to write a general code for all types of images(targets). I don't know how use a monocular plane detection and yes, these images are part of a video stream. Thanks for all informations. – Leonard993 Oct 06 '18 at 13:49
  • Ah cool, if they are part of a video stream there is a lot of info! I think that one good place to start would be to manually initialize the bounding box of your target, and then use a Kalman filter - or particle filter if you don't need to be real time, it's easier to implement - (see https://docs.ufpr.br/~danielsantos/ProbabilisticRobotics.pdf for a gentle intro to those) to track the evolution of said bounding box. That being said, I don't think that a generic code to track all type of objects is a reasonable objective for you right now (this is more or less an active area of research). – Ash Oct 06 '18 at 14:05

0 Answers0