Prev: How to combine volume and surface area calculations for adjacentsets of data
Next: Internet connection issue
From: Conor on 19 May 2010 04:55
It might make it easier to explain with an image.

http://i3.photobucket.com/albums/y60/con20or/Target.png

The x/y blue points are the ones that Im trying to fit the in the target-circles around.

The x/y white points are ones i dont want to fit. In practice they would be an approximately m more even distribution, but to illustrate the fact that I could be left with quite a small amount of points to fit inside the target circles, ive deleted them.

I want to take a set shape (the cross with circular targets on the end) and make sure that it is placed so that each circle only contains blue points, not white ones. I then want to be able to check the x/y value of the cross centre and compare it against a known x/y value.

If I had only used one circle, it could have fit numerous ways around the blue dots (there could be even less than is shown in this image, clustered around each other. When i checked it manually in CAD, two was better, three better again, four marginally better than three.
From: Bruno Luong on 19 May 2010 05:28
"Conor " <remove.this-con20or(a)gmail.com> wrote in message <ht091c$qaq$1(a)fred.mathworks.com>...
> It might make it easier to explain with an image.
>
> http://i3.photobucket.com/albums/y60/con20or/Target.png
>
> The x/y blue points are the ones that Im trying to fit the in the target-circles around.

The problem you have might be call CLUSTERING, but not FITTING. FITTING has entirely different meaning for many of us.

Take a look at K-mean clustering algorithm for example.

Bruno
From: David Young on 19 May 2010 06:33
Would the following make sense:

Set up an array of zeros, and some suitable mapping from array indices to positions in the space of points. Set array elements at blue dots' positions to +1 and those at white dots' positions to -1 (or maybe some other negative number, reflecting the cost of having white dots inside the circles).

Set up another array of zeros, set elements within the four circles to +1.

Use cross-correlation (or convolution) to find the offset at which the largest number of blue dots fall in the circles.

There's a trade off between the accuracy of the result and the memory and time needed.
First  |  Prev  | 
Pages: 1 2
Prev: How to combine volume and surface area calculations for adjacentsets of data
Next: Internet connection issue