0

With a camera inside a cylinder I capture a image. I want to detec if there are some deformation due to a collision outside. I also want to detec in which side the collision occurs. The image inside the cylinder have a lot of dots which forms a grid. Which is the better way to do this?

enter image description here

A simple way to detec the collision is to subtract the image without collision with the real image. If the result isn't "zero", something changed and probally a collision occured. But this doesn't give me which side the cylinder deformed.

I already tried to do a projection of the points in the plane, but i couldn't do it.

In this link you can find a question post by me with the problem of the projection: Projection of a image from inside a cylinder to a plane 2D [Matlab]

In that link you can see all the information about this problem.

An idea is to use region props in the image and see which part of the image deformed, but I want to do something a little more complex. I want to measure the deformation, to have an idea how much it deformed during the impact. This is the reason why i thought about doing some projection in the plane and measure the distance that the points deformed. Do you have any idea to do this in a more simple way? How can i do it

Someone can help me please?

Community
  • 1
  • 1
Afonso
  • 46
  • 8

1 Answers1

1

Here's a little code/pseudo-code to try to help. In words:

I would subtract the before and after images and take the absolute value of the difference image. Then, I would have some sort of threshold for whether or not the difference is just due to variation in noise and not a real change. Next I find the center of mass (weighted by the magnitude of difference), which can be done easily with the image processing toolbox (regionprops). The center of mass of the variation would be a good estimate of where a "collision" occurred, i.e. a deformation in the cylinder

So that would be something along the lines of:

diffIm = abs(originalIm - afterIm)
threshold = someNumber
diffIm = diffIm(diffIm>threshold)

%Tell regionprops that the whole image is one region by passing it a array of ones the size of the image, and diffIm as the measurment image
props = regionprops(ones(size(diffIm)),diffIm,'WeightedCentroid')
%WeightedCentroid is the center of mass, and it is weighted by the grayscale image diffIm

You now have the location of the centroid of deformation in your image space, and all you would need is a map to convert that to cylinder space (if you needed that), otherwise you could just plot the centroid over the original image for a visual output of where the code expects the collision occurred.

On another note, if you have control of your expirimental setup, I would expect that a checkerboard pattern would give you better results than the dots (because the dots are very spaced out, and if the collison only affects the white space you might not be able to detect it at all). A checkerboard would mean you have more edges than can be displaced, which is the brunt of what would be detected anyways. A checkerboard may also be easier for mapping to a plane if you were still trying to do that, because you could know that all the edges are either parallel or intersecting at right angles, and also evenly spaced.

Tar
  • 323
  • 1
  • 11
  • Sorry to answer so late, i had some troubles during this week. That's a great idea and it should work. But I want to do something a little more complex. I want to measure the deformation, to have an idea how much it deformed during the impact. This is the reason why i thought about doing some projection in the plane and measure the distance that the points deformed. Do you have any idea to do this in a more simple way? How can i do it? – Afonso Sep 17 '15 at 09:40