Recursive Algorithm
Recursive filters provide an optimal estimate of the internal state of a linear system from a series of noisy measurements. Implementing kalman state estimation involves a two-step process consisting of a prediction phase and a correction phase.
Error Covariance
Matrix algebra tracks the uncertainty of the system state through a transition model and a measurement model. In kalman state estimation, the gain factor determines the weight assigned to new sensor data relative to the predicted value. High noise levels in the sensor input lead to a lower gain and a greater reliance on the model.
Realtime Feedback
Low computational overhead allows for continuous updates in embedded hardware environments. Applications for kalman state estimation include sensor fusion where data from accelerometers and gyroscopes combine to determine orientation. The filter removes high frequency noise while maintaining the response speed of the system.
Linear Model
Performance degrades when the underlying system dynamics exhibit strong nonlinearities. Standard kalman state estimation assumes that both the process noise and the measurement noise follow a Gaussian distribution. Extended versions of the algorithm linearize the system around the current estimate to handle non-linear transitions by calculating the linearization matrix at every time step.