For the next 3 questions, let us assume that we have a senso…

For the next 3 questions, let us assume that we have a sensor collecting a 3D point cloud in an environment. From this data, we need to design an algorithm to identify generic infinite cylinders: the main axis of the cylinder and its radius are unknown. However, we do not try to recover the length of the cylinder (it is infinite). Note that a generic cylinder is completely constrained with 4 points.

In vision, it is very common to solve linear systems of the…

In vision, it is very common to solve linear systems of the form A · X = 0, with A ∈ IRn×m  contains known values and X ∈ IRm is a vector of parameters, with n >> m. They are solved by minimizing ||A · X|| such that ||X|| = 1. Why not use a standard regression formula with the pseudo-inverse instead?