I have a pretty noisy Nx2 data, where N is time stamp and the two columns are x, y position in a 2D plane.  The curve could look like this (upper-left panel)
http://www.pnas.org/content/103/18/7192/F1.large.jpg 
or, here is a much less noisy example
http://live.ece.utexas.edu/research/doves/DovesImages/sub1.jpg

I wonder what's the algorithm that will find several line segments that best fit this kind of crazy curve, and is there a matlab code available to do it?  thanks

On Jan 21, 9:17=A0pm, "Po-He" <pohe...@gmail.com> wrote:
> I have a pretty noisy Nx2 data, where N is time stamp and the two columns=
 are x, y position in a 2D plane. =A0The curve could look like this (upper-=
left panel)http://www.pnas.org/content/103/18/7192/F1.large.jpg
> or, here is a much less noisy examplehttp://live.ece.utexas.edu/research/=
doves/DovesImages/sub1.jpg
>
> I wonder what's the algorithm that will find several line segments that b=
est fit this kind of crazy curve, and is there a matlab code available to d=
o it? =A0thanks

---------------------------------------------------------------------------=
------
Looks like you already have a curve - the blue line.  Do you want to
fit a simpler, less wiggly curve?  Let's say that you could do that.
Then what?  What are you *really* after such that you think you need a
fitted curve?  What will a fitted curve do for you, assuming you had
one?  In other words, what's the larger picture - the context - here?
Can you share that, in case there's a way to get that without doing
some kind of curve fitting?

You might be interested in Douglas-Peucker line simplification algorithm. It is *not* a fit algorithm, just a method of selecting a subset of vertexes in the original curve such that the piecewise linear curve is well approximate the original data.

There are few such implementations on FEX.

Bruno

Thanks for the reply.  In the end, I'd like to know "when" and "how much" the line change its direction.  Therefore, a curvy turn is hard to decide the exact angle and the timing.  therefore, I'd like to fit these curves with a few straight line.  therefore, it's easier to calculate the angular change and the timming of direction change.  thank you

pohe

ImageAnalyst <imageanalyst@mailinator.com> wrote in message <6e9e527d-d472-4d34-8646-98c94b51170d@c39g2000yqi.googlegroups.com>...
> On Jan 21, 9:17 pm, "Po-He" <pohe...@gmail.com> wrote:
> > I have a pretty noisy Nx2 data, where N is time stamp and the two columns are x, y position in a 2D plane.  The curve could look like this (upper-left panel)http://www.pnas.org/content/103/18/7192/F1.large.jpg
> > or, here is a much less noisy examplehttp://live.ece.utexas.edu/research/doves/DovesImages/sub1.jpg
> >
> > I wonder what's the algorithm that will find several line segments that best fit this kind of crazy curve, and is there a matlab code available to do it?  thanks
> 
> ---------------------------------------------------------------------------------
> Looks like you already have a curve - the blue line.  Do you want to
> fit a simpler, less wiggly curve?  Let's say that you could do that.
> Then what?  What are you *really* after such that you think you need a
> fitted curve?  What will a fitted curve do for you, assuming you had
> one?  In other words, what's the larger picture - the context - here?
> Can you share that, in case there's a way to get that without doing
> some kind of curve fitting?

cool, that's exactly the thing I was looking for.  thanks for the help!!

pohe

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <ihe4as$sjq$1@fred.mathworks.com>...
> You might be interested in Douglas-Peucker line simplification algorithm. It is *not* a fit algorithm, just a method of selecting a subset of vertexes in the original curve such that the piecewise linear curve is well approximate the original data.
> 
> There are few such implementations on FEX.
> 
> Bruno

On Jan 22, 4:23=A0pm, "Po-He" <pohe...@gmail.com> wrote:
> Thanks for the reply. =A0In the end, I'd like to know "when" and "how muc=
h" the line change its direction. =A0Therefore, a curvy turn is hard to dec=
ide the exact angle and the timing. =A0therefore, I'd like to fit these cur=
ves with a few straight line. =A0therefore, it's easier to calculate the an=
gular change and the timming of direction change. =A0thank you
>
> pohe
-------------------------------------------------------------------------
pohe:
Maybe you should look at the fractal dimension of the curve.  That
should be a good metric for characterizing how tortuous the curve is
and how completely it fills the plane.
ImageAnalyst

Thanks for the suggestion.  I'll look into that.

pohe

ImageAnalyst <imageanalyst@mailinator.com> wrote in message <5cf5be2b-27d4-450a-946a-9f6121f59953@k42g2000yqa.googlegroups.com>...
> On Jan 22, 4:23 pm, "Po-He" <pohe...@gmail.com> wrote:
> > Thanks for the reply.  In the end, I'd like to know "when" and "how much" the line change its direction.  Therefore, a curvy turn is hard to decide the exact angle and the timing.  therefore, I'd like to fit these curves with a few straight line.  therefore, it's easier to calculate the angular change and the timming of direction change.  thank you
> >
> > pohe
> -------------------------------------------------------------------------
> pohe:
> Maybe you should look at the fractal dimension of the curve.  That
> should be a good metric for characterizing how tortuous the curve is
> and how completely it fills the plane.
> ImageAnalyst