Eight-Point Algorithm: Finding Geometry from Image Pairs

WHY IT EXISTS: In computer vision, if you have two images of the same scene from different viewpoints (a stereo pair), you need a way to mathematically describe the geometric relationship between the cameras. This relationship is crucial for 3D reconstruction. The eight-point algorithm was created to solve this by estimating that geometry from the images themselves.

THE MENTAL MODEL: Think of it as a geometric detective. You give it at least eight pairs of 'clues'—points that you know correspond to the same real-world location in both images. The algorithm uses these clues to deduce the fundamental geometric relationship, encapsulated in a special matrix (the essential or fundamental matrix), that links the two camera views.

HOW IT WORKS: The algorithm takes a set of corresponding image points from a stereo camera pair as its input. It uses these points to formulate and solve a system of linear equations, which determines the unknown values of the essential or fundamental matrix. The requirement of at least eight points is the minimum needed to find a unique solution for this system.

WHEN TO USE IT: Use the original eight-point algorithm, as introduced by Christopher Longuet-Higgins in 1981, when you need to estimate the essential matrix from a set of corresponding image points. This is its primary and intended application.

WHEN NOT TO USE IT: Avoid using the original algorithm to compute the fundamental matrix in a practical setting. While theoretically possible, it is numerically unstable. For estimating the fundamental matrix, the 'normalized eight-point algorithm' described by Richard Hartley in 1997 is the robust and preferred method.

ONE CANONICAL EXAMPLE: The classic application is estimating the essential matrix for a stereo camera pair. Given two images and at least eight matching feature points between them, the algorithm computes the essential matrix, which encodes the relative rotation and translation between the two camera positions.