What is optical flow?

The world streaming past

Look out of a train window: nearby fences rush by, distant hills crawl. That apparent motion of the scene across your eye, or across a camera’s image, is optical flow. Measured at many points, it becomes a field of little arrows showing which way each part of the picture moved, and how far.

Measuring it is harder than it sounds. As Horn and Schunck pointed out in 1981, the brightness at a single point gives one measurement, but motion in a picture has two directions, so a method needs an extra assumption. Theirs was that the flow changes smoothly across most of the image. Many later methods build on that idea.

From flow to speed

A camera looking straight down sees the floor move. The same speed makes the floor race past at 20 cm and drift past at 2 m, so the drone needs its height, from a distance sensor, to turn image motion into metres per second. It also has to subtract its own tilting: when a drone leans to start moving, the floor appears to move before the drone does, and a gyroscope measures that rotation so it can be taken out.

A 2013 open-source, open-hardware design by Honegger and colleagues put the whole recipe on one small board: a light-sensitive camera chip, flow computed on a microcontroller 250 times a second, rotation removed with a gyroscope and distance from an ultrasonic sensor, shown in flight on a small drone. The authors noted that earlier flow sensors built from computer-mouse chips needed more light than typical indoor and outdoor low-light conditions provide.

Bees did it long before

How does a honeybee know how far away the flowers are, so that its dance can tell the hive? One theory said it measures the energy it spends; another, that it judges how much the world has moved across its eyes. Srinivasan and colleagues tested the two by letting bees fly through a short, narrow tunnel to reach food, and watching their dances afterwards. The result, published in 2000: the bee’s “odometer” is driven by vision.

An optical mouse uses the same principle on a smaller scale: a tiny camera underneath watches the desk surface move.

Where it fails

• Blank or shiny floors. A plain, glossy surface gives the camera little to track.
• Low light. Fewer details, more noise.
• A wrong height. Every error in the distance reading becomes an error in speed.
• Drift. Adding up speed to get position also adds up its small errors, so a drone that relies on flow alone slowly wanders. Robots correct this against a map: what SLAM is.

How Alkyon is designed

Alkyon, our concept for a flying home companion, is designed with a downward optical flow sensor beside a downward depth sensor with 64 zones, so it can tell both how fast it moves and how high it is. In our simulation we model the errors such sensors make, and the drift they add up to is one of the main open problems for flying close to furniture. It is a concept in development, for Europe only: how indoor drones navigate.

Sources, read 4 November 2026: Horn and Schunck, determining optical flow, Artificial Intelligence, 1981; Srinivasan and colleagues, honeybee navigation: nature and calibration of the “odometer”, Science, 2000; Honegger and colleagues, an open source and open hardware embedded metric optical flow CMOS camera for indoor and outdoor applications, IEEE ICRA, 2013. General information.