What Is Optical Flow in Computer Vision?

Optical Flow in Computer Vision Explainer

Optical flow is an estimate of apparent motion between two images, usually consecutive video frames. A dense flow field assigns a horizontal and vertical displacement to each pixel, while sparse methods track selected points.

Classical algorithms rely on assumptions such as brightness consistency and locally smooth motion. Modern neural systems learn motion patterns from labeled or self-supervised data, often combining multi-scale features and iterative refinement.

Flow supports video stabilization, frame interpolation, action analysis, object tracking, robot navigation, and motion-aware compression. It describes how visible image patterns move, which is not always identical to the physical motion of objects in 3D.

Occlusion creates pixels with no true match in the other frame. Reflections, shadows, fast motion, blur, textureless surfaces, and repeated patterns also cause ambiguity. Evaluation should include realistic motion, boundaries, changing illumination, and confidence estimates.

Optical flow estimates how image patterns move from one video frame to the next, usually as a motion vector for each pixel.

Key terms include optical flow, flow field, displacement vector, brightness constancy, smoothness, occlusion, motion boundary and correspondence.

The flow field links image locations across nearby frames.

Multi-scale reasoning helps capture both small movements and large displacements.

Occlusions and textureless regions may have no unique correspondence.

Lighting changes and reflections can move visually even when objects do not.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.