Kalman filters are usually used to estimate the 'states' of a system.
The states of a system are defined as the information needed to predict the future response of the system given the future inputs. Many times it is not possible or too expensive to measure all of the system states, or they are corrupted by noise.
Hence they must be estimated. The kalman filter is an optimal estimator, which uses the measured outputs, and process inputs to estimate the states. IT is useful because it allows a recursive solution.