Mon Jun 30 14:20:24 CEST 2008

2D direction estimation (confused)

I first thought this would be Hough transform in r,t with the radial
direction integrated out. However, that doesn't work because of the
way accumulation works: each t is voted for in the same way, but with
different r.

I can't see how a bundle-accumulator would work for estimating
direction. How can a point vote for a direction? Doesn't make sense.

What about hierarchical Hough transform? We know there are only 2
directions, and they are orthogonal, so basically, there is only
one. This looks like a successive binary estimation problem. First
choice between t=0,pi/4

But, doing it this way: how to pick one of the two angles? The result
will be 2 accumulators for an r spectrum. We're not interested in r,
but it will look like a spectrum with multiple (blunt) peaks. The
decision rule could be something like high-pass content (local