Residual Uncertainty
Statistical measures of the difference between a predicted sensor measurement and the actual observation provide a metric for the health of a tracking filter. The innovation covariance quantifies the expected variance of this error. It combines the uncertainty of the system model with the noise of the sensor itself.
Computation Step
The matrix is calculated by projecting the state uncertainty into the measurement space. In the kalman filter, innovation covariance is used to determine the weight given to new data. A high covariance indicates that the new measurement is less reliable.
Outlier Detection
Engineers monitor the innovation covariance to identify sensor failures or unexpected environment changes. If the actual residual is much larger than the predicted innovation covariance, the system might be experiencing a fault. This comparison is the basis for statistical gating techniques.
The matrix helps in setting the boundaries for what the filter considers a normal observation. Consistent deviations suggest that the sensor noise model or the system dynamics need adjustment. This diagnostic value makes it an essential parameter for autonomous navigation.
Filter Gain
The kalman gain is inversely proportional to this matrix. Therefore, innovation covariance directly influences how much the system updates its internal state based on new information.
