I am developing a project where I need to track moles skin. I already succeeded using color tracking, but I would need something more precise.
I was thinking about using ir ink and paint the moles. Have you guys ever worked with IR ink or similar? Do you have another better idea for achieving moles tracking. What camera-lenses combination should I use?
Thanks you all in advance!
What else will be in the image? If it is a relatively plain image you could do simple thresholding; then use OpenCV or similar to find blobs. That would be less sensitive to light and color change then color tracking would be, I think, since color tracking relies heavily on a particular color to remain visible. However you would need first to isolate the area where thresholding is posible.
Thank you mennowo,
I will give it a try doing just a simple thresholding.
Here is a first attempt to achieve the moles skin tracking thing (I had to paint the moles): https://vimeo.com/87025438
For this first attempt, the camera is very close to the skin, so the moles are bigger, but in the future I would like to put the camera far from the skin (moles would be smaller) and that is why I am trying to find something more precise, because tracking blob would be much smaller.
BTW, the idea of isolate the area where thresholding is posible is a good one, and I think it can improve tracking a lot.
Looks nice! I like the soundscape, what software/language did you create it with?
With your setup I think thresholding needs to be somewhat more elaborate; the surrounding area is dark, so it would be recognized as something you are looking for if the moles are also dark.
Maybe you can first find in the image the (relatively) large area where the skin is visible; it will all (except for the small moles) be quite light; maybe edge detection or other means to isolate that area, including dark spots within it; thus create a histogram and then use that to ignore the surroundings and only scan the found ROI for moles. I guess you could also try color detection on the ROI.
Then I think blob tracking can be quite precise, especially if the contrast between the moles and the skin is rather large.
In OpenCV you can specify how big blobs need to be at least to be seen as such. Maybe you can also specify a maximum size, or you analyze blobs found by yourself and then ignore large ones; then maybe even you do not need to isolate anything; just make a histogram by thresholding, and find blobs in it. If they are small, they must be moles. Provided the background will not have light colors and disrupt this by providing false “moles”.
See this page: http://opencv-users.1802565.n2.nabble.com/Image-segmentation-using-Blob-Detection-in-Open-CV-td7583380.html.
Then this link, rather old but maybe helpful for your purposes: http://www.shervinemami.info/blobs.html.
Thanks a lot mennowo,
it makes sense, I will try that.
I use supercollider for generating sound, which is modulated by openframeworks via OSC, depending on the size of the blob and the distance between them.