The Kalman filter is an algorithm mainly (but not only) developed by Rudolf Kalman in the late 1950s, and has found practical application in many fields, particularly navigation systems for all sorts of vehicles from nuclear submarines to aircrafts.

The Kalman filter operates recursively on streams of noisy input data (which in computer vision is normally a video feed) to produce a statistically optimal estimate of the underlying system state (the position inside the video).

Let's take a quick example to conceptualize the Kalman filter and translate the preceding (purposely broad and generic) definition into plainer English. Think of a small red ball on a table, and imagine you have a camera pointing at the scene. You identify the ball as the subject to be tracked, and flick it with your fingers. The ball will start rolling on the table, following the laws of motion we're familiar with.

If the ball is rolling at a speed of 1 meter per second (1 m/s) in a particular direction...