An edge is a derivative. A camera is division by depth. Stereo, optical flow, and surface normals are the same calculus, just aimed at recovering the missing Z.
Learning Objectives
→Define an edge as the magnitude of a finite-difference image gradient, and read its orientation with atan2
→Project a 3D point through a pinhole camera and back-project a pixel once depth is known
→Build a unit surface normal as the cross product of two surface tangents
→Recover depth from stereo disparity and quantify how that estimate blows up as disparity shrinks
→Solve the brightness-constancy constraint for the flow component along the image gradient, then lift a depth map into a point cloud