difference between plane segmentation and plane fi

Question

I've recently been working on a project in wich I'd have to detect walls, floor and ceiling in a 3D mesh. After doing some research I've been able to detect the floor and some part of the walls using RANSAC algorithms. I was just wondering if anybody could be able to explain the difference between plane-fitting and plane segmentation as they both seem to result into a point cloud containing the floor?

Answer 1

Plane fitting is generally understood as a pure least-squares based fitting technique where distance from a point set to the equation of a plane is minimized. One could formulate such optimization in many ways, but the most practical one is a simple SVD analysis of the covariance matrix.

Plane segmentation can be thought more as an instance of the fitting problem where a lot of outliers exist in the data. In fact, the plane to segment could be a tiny fraction of the point set. Moreover, there could be multiple planes in different orientations. Finally, the segmentation problem might not necessarily search for the plane equations but rather concerns finding out the points, which are jointly lying on a plane or multiple sets of points lying on multiple planes. Though, many algorithms which aim to do that implicitly use a fitting technique with some geometric or algebraic outlier handling such as RANSAC, Hough Transform, robust norms etc.

It is also important to understand that plane detection can be done only up to a certain accuracy. Figuring out which points lie on a particular plane is a different problem and sometimes is more complex than just taking out a bunch of points which are close enough to the plane.

For the case of identifying walls, ceilings and etc. both fitting and segmentation techniques will be necessary.