1 | initial version |

Probably the radial transform is not the best way to detect ellipses.

From the definition of the transform, every pixel affects only the pixels along the gradient vector in the current pixel. This is perfect to detect circles, as the radius vector is always perpendicular to the contour and the distance to the center of the circle is constant (=radius).

However in the case of ellipses, the contour isn't perpendicular to the radius, and the radius varies also; so it won't give a single bright pixel in the accumulator space.

Maybe with some modifications of the original method you could obtain some results, but I think there are better suited algorithms for ellipse detection.

2 | No.2 Revision |

Probably the *radial transform transform* (Loy&Zelinsky) is not the best way to detect ellipses.

From the definition of the transform, every pixel affects only the pixels along the gradient vector in the current ~~pixel. ~~pixel and to a given distance `n`

. This is perfect to detect circles, as the radius vector is always perpendicular to the contour and the distance to the center of the circle is constant (=radius).

However in the case of ellipses, the contour isn't perpendicular to the radius, and the radius varies also; so it won't give a single bright pixel in the accumulator space.

Maybe with some ~~modifications of the original method ~~adaptations you could obtain some ~~results, ~~results (by varying the `sigma`

parameter) , but I think there are better suited algorithms for ellipse detection.

Copyright OpenCV foundation, 2012-2018. Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.