Optical flow is the pattern of apparent motion between video frames — the per-pixel arrows showing which way things moved. It lets robots and drones sense their own motion, track objects, and detect what's moving around them.
Optical flow measures how each part of an image moves from one video frame to the next — a field of little arrows. A drone uses it to tell how fast it's sliding sideways; a robot uses it to spot moving objects.
Watch the ground rush past from a moving car and you instantly sense your speed and direction. A robot camera can extract the same cue from video: it's called optical flow.
What it is
Optical flow is the field of apparent motion between two consecutive frames — for each pixel (or feature), an arrow showing which way and how fast it moved. It's computed from how brightness patterns shift, assuming a point's brightness stays roughly constant as it moves (the brightness-constancy assumption).
From two frames to a motion field
Compare where each patch of brightness went between frames and you get a vector per pixel — the optical flow field.
Sparse vs dense
Sparse flow (Lucas-Kanade) tracks a handful of strong feature points — fast, robust, and enough for visual odometry or object tracking.
Dense flow (Horn-Schunck, and modern networks like RAFT) computes a vector for every pixel — richer, used for segmentation of moving objects, video stabilization, and depth-from-motion.
What robots do with it
Ego-motion / speed sensing. A downward camera on a drone reads flow to estimate sideways drift and hold position indoors where there's no GPS — the trick behind consumer drones' rock-steady hover.
Moving-object detection. Subtract the flow caused by the robot's own motion and whatever's left is something else moving — a pedestrian, another vehicle.
Tracking and visual servoing. Following features frame-to-frame.
The limits
Flow breaks where the brightness-constancy assumption does: sudden lighting changes, reflections, and large fast motions. It also suffers the aperture problem — looking through a small window at a moving edge, you can only tell motion across the edge, not along it — which is why flow needs texture and corners, not blank walls.
Why it matters
Optical flow is one of the oldest and most useful motion cues in vision — the bridge between static images and understanding movement, and a building block of visual odometry, tracking, and drone stabilization.
