I’m trying to detect some small attachments to the human body where the person wears minimal cloth. So, these attachments are directly on the skin.
I’ve tried color filtering to reveal these small attachments but it really did not work since color of the attachments are not too far away from what is present in the image
Also, Hough circles extracted too many circles.
Another thing, I tried is feature extraction by SIFT. It extracts too many features that are hard to choose from.
How can I detect the position of the attachments as shown in the example image below:
The people who made this picture are doctors of some field. Of course they have no knowledge of how things work in image processing and/or computer science.
They asked me whether this setup can be used to detect certain joints on the human body.
I myself tried some general things but could not detect the circles.
So, do you think it is not entirely possible or can be possible to an extend or the setup should be updated with something else?
they won’t get a satisfactory answer from you because you’re here asking.
“detect circles” tells me you don’t have an intuition for what’s going on.
these balls are used for things like “motion capture”. if used correctly, detecting them is trivial. they aren’t used correctly here.
“detect joints” is too vague. these medical people need to explain the goal.
I can only repeat myself. you need to get a comprehensive education on image processing and computer vision, or defer the task to someone else. question-answer rounds won’t do.
you should research these things. don’t expect people on a forum to give you one-on-one lessons for things you could simply get from existing material (lectures, books, journals, blogs, youtube videos, company websites, … etc).
First of all, thank you for answering properly. I was trying to understand whether these balls can be captured properly with high accuracy within the current settings or not. I can understand your answer is almost a “no”.
Secondly, beyond answering the question, judging me and my background is a bit rude, I have to tell you. You could just answer by showing the resources to be read or give no answer at all. If you are pissed off by anyone asking the easiest things here or in contrast the hardest things without any clue of the thing at all, then just close the question and tell them “This area is for experienced people”. I would understand. But being (kind of) insulted for just asking a question is nonsense.
Yes, you can detect them this way with a lot of work.
A better way, as crackwitz pointed out, is to use a fiducial marker easy to detect on the image, like markers with bright color, with light or reflecting ones actually reflecting bright light to the camera. So you can easily segmentate them by brightness or color.
By your answers, I am convinced that the setup should be changed to something else in which the “markers” can be more easily distinguished from the background by one of the following such as brightness, color or shape.
I guess the following one closer to what you’ve been talking about:
there is no reason to involve color at all. that’s a waste of precious light and sensor area. it’s doable, but it’s far from good.
the point of retroreflectors is that they’ll be brighter than everything else in the picture, because a light near the camera illuminates them and all that light comes back to the camera. that means detecting them is just thresholding. no other processing at all. the camera’s exposure settings would be set such that everything else is fairly dark and the reflectors aren’t “bleeding light” too much.
I’d expect that (now) knowing the keywords “motion capture” and retroreflector enables research into the “state of the art”, or at least proven methods.
