Homogeneous Coordinates: A Lecture by Cyrill Stachniss (2020)

Added:

Conformal Geometric Algebra (CGA)

While homogeneous coordinates and projective geometry are the industry standards in computer vision and robotics, Conformal Geometric Algebra (CGA) is increasingly advocated as a more powerful and mathematically unified alternative. Homogeneous coordinates represent projective space by adding an extra dimension, but they rely heavily on matrix algebra, which can obscure geometric intuition and struggle to natively represent entities like circles and spheres. In contrast, CGA (a branch of Clifford Algebra) represents points, lines, planes, circles, and spheres as direct geometric objects (blades) within a 5D space. Instead of using $4 \times 4$ transformation matrices, CGA utilizes 'rotors' to perform rotations, translations, and scaling through unified sandwich products. This coordinate-free approach avoids the numerical instability of perspective division inherent to homogeneous coordinates, simplifies the intersection of geometric primitives, and provides a more elegant, computationally robust framework for spatial reasoning in robotics and computer graphics.

Basic linear algebra, including vector operations, matrix-vector multiplication, and matrix inversion.
Cartesian coordinate systems in 2D and 3D space.
Standard geometric transformations such as rotation, translation, and scaling in Euclidean space.
The mathematical distinction between linear transformations and affine transformations.
The pinhole camera model and the use of projection matrices in computer vision.
Epipolar geometry, including the estimation of Essential and Fundamental matrices for stereo vision.
Homography estimation and its application in image stitching and planar perspective rectification.
3D reconstruction techniques, such as Structure from Motion (SfM) and photogrammetry workflows.
Robot kinematics and coordinate frame transformations for robotic manipulation and SLAM.
32.2K views824likes1:10:19@CyrillStachnissOriginal Release: 2020-08-20

Homogeneous coordinates provide a mathematical framework for representing geometric objects in projective geometry, where points and lines are defined up to a scalar factor, enabling elegant representation of points at infinity (where the last coordinate equals zero) and allowing all transformations (translation, rotation, scaling, affine, and projective) to be expressed through matrix-vector multiplication, making complex geometric operations like line-point intersection and line fitting computationally efficient.