reduce time taken for ellipse detection

[Stéfan van der Walt]

Hi Austin

On Tue, Jun 17, 2014 at 8:47 PM, Austin Chungath <austincv at gmail.com> wrote:
> I am trying out the ellipse hough transform as described in the following
> link:
> http://scikit-image.org/docs/dev/auto_examples/plot_circular_elliptical_hough_transform.html#example-plot-circular-elliptical-hough-transform-py
>
> The ellipse detection in the example above with the coffee mug takes about
> 15 seconds to give a result.
> Unfortunately, I need to be able to at least do this in under a second since
> I am working on a video stream.

This algorithm does a type of brute-force search, so I don't think it
will ever be fast enough.  A probabilistic approach (as in the
probabilistic Hough line transform we implemented) would be better.
You can also scan the mailing list for a discussion we had a while ago
on fast detection of circles, which may be extended to ellipses.

Dan's suggestion of looking at OpenCV is also a good one; I am not
familiar with all the algorithms they provide, but it will undoubtedly
be fast.

Regards
Stéfan