Kalman filter

A recursive predict-then-correct estimator that tracks an object's state frame-by-frame under uncertainty.

A Kalman filter recursively estimates a tracked object's state (position, velocity) frame by frame under uncertainty, alternating a predict step (project the state forward using a motion model) with a correct step (blend that prediction with a new noisy measurement, weighted by each source's own uncertainty) to produce a smoother, more accurate estimate than either alone.

Examples

Tracking a thrown ball, a Kalman filter predicts its next position from its known velocity, then corrects that prediction against the next frame's noisy detected position.