Hough Flashcards
Line fitting challenges
Extra feature points (clutter)
Only some parts of a line are detected
Noise in measured feature points
Voting is
A technique where we let each feature vote for all the models that are compatible with it
Hough transform
Global shape parametrized by line or circle,
Edge element votes for particular parameter values
Global shape detected by many votes for same parameter value
Hough transform outline
Separate images into bins
For each feature point in image, vote in every bin that could have generated this point
Find the bins with the most votes
A line in an image corresponds to a
Point in hough space
A line is a set of pints such that
X cos theta + y sin theta = distance
Theta is angle of perpendicular line to the x axis
Hough transform algorithm
H(theta,d) =0
For each edge point (x,y)
For min theta to max theta
D = x cos theta + y sin theta
H(theta, d) += 1
Find the max if H(theta, d)
Dealing with noise in hough transform
Take points only with significant gradient magnitude
Choose grid bin size carefully
Hough transform for circles
Alpha = a,b,r
For each edge point [x, y, mag, dir]woth mag > t
For each possible radius value r Theta =gradient orientation at x,y A = x-r cos theta B= y-r sin theta H(a,b,r) += 1
Hough pros
Can cope with occulusion
Can detect multiple instances of a model
Robustness to noise
Hough cons
Complexity of search time increases exponentially with number of model parameters
Non target shapes can produce spurious peaks in parameter space
Hard to pick good grid size