You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Implement Moment of Inertia-based Crescent Edge Elimination on page 12. If the planet shows up as a crescent shape, this eliminates the edge points that are inside the planet but show as a crescent edge.
Tip
This is an eigenvalue problem, but its very trivial (2x2 matrix). For any 2x2 matrix A, the eigenvalue and eigenvector you're looking for is: $$\lambda_{max} = \frac{1}{2}\left(tr(A)+\sqrt{tr(A)^2-4det(A)}\right)$$ $$\vec{w} = [1,\ (\lambda - a)/b ]\ OR\ [(\lambda-d)/c,\ 1] $$
Note
This algorithm operates by finding the long axis of symmetry and then doing simple fits on both sets of points to figure out which one has less outliers. You will notice this sounds a lot like RANSAC. You should find a way to share the code that RANSAC has (by pulling out a utility function from RANSAC) and use that to your advantage.
Description
Implement Moment of Inertia-based Crescent Edge Elimination on page 12. If the planet shows up as a crescent shape, this eliminates the edge points that are inside the planet but show as a crescent edge.
Tip
This is an eigenvalue problem, but its very trivial (2x2 matrix). For any 2x2 matrix A, the eigenvalue and eigenvector you're looking for is:
$$\lambda_{max} = \frac{1}{2}\left(tr(A)+\sqrt{tr(A)^2-4det(A)}\right)$$
$$\vec{w} = [1,\ (\lambda - a)/b ]\ OR\ [(\lambda-d)/c,\ 1] $$
Note
This algorithm operates by finding the long axis of symmetry and then doing simple fits on both sets of points to figure out which one has less outliers. You will notice this sounds a lot like RANSAC. You should find a way to share the code that RANSAC has (by pulling out a utility function from RANSAC) and use that to your advantage.