Trait ObservationModel 

pub trait ObservationModel<R, SS, OS>
where
    R: RealField,
    SS: Dim,
    OS: Dim + DimMin<OS, Output = OS>,
    DefaultAllocator: Allocator<SS, SS> + Allocator<SS> + Allocator<OS, SS> + Allocator<SS, OS> + Allocator<OS, OS> + Allocator<OS>,
    Matrix<R, SS, SS, Owned<R, SS, SS>>: One,
{
    // Required methods
    fn H(&self) -> &Matrix<R, OS, SS, Owned<R, OS, SS>>;
    fn HT(&self) -> &Matrix<R, SS, OS, Owned<R, SS, OS>>;
    fn R(&self) -> &Matrix<R, OS, OS, Owned<R, OS, OS>>;

    // Provided methods
    fn predict_observation(
        &self,
        state: &Vector<R, SS, Owned<R, SS>>,
    ) -> Vector<R, OS, Owned<R, OS>> { ... }
    fn update(
        &self,
        prior: &StateAndCovariance<R, SS>,
        observation: &Vector<R, OS, Owned<R, OS>>,
        covariance_method: CovarianceUpdateMethod,
    ) -> Result<StateAndCovariance<R, SS>, Error> { ... }
}

An observation model, potentially non-linear.

To use a non-linear observation model, the non-linear model must be linearized (e.g. using the prior state estimate) and use this linearization as the basis for a ObservationModel implementation. This would be done every timestep. For an example, see nonlinear_observation.rs.

Required Methods

fn H(&self) -> &Matrix<R, OS, SS, Owned<R, OS, SS>>

Get the observation matrix, H.

fn HT(&self) -> &Matrix<R, SS, OS, Owned<R, SS, OS>>

Get the transpose of the observation matrix, HT.

fn R(&self) -> &Matrix<R, OS, OS, Owned<R, OS, OS>>

Get the observation noise covariance, R.

Provided Methods

fn predict_observation( &self, state: &Vector<R, SS, Owned<R, SS>>, ) -> Vector<R, OS, Owned<R, OS>>

For a given state, predict the observation.

The default implementation implements a linear observation model, namely y = Hx where y is the predicted observation, H is the observation matrix, and x is the state. For a non-linear observation model, any implementation of this trait should provide an implementation of this method.

If an observation is not possible, this returns NaN values. (This happens, for example, when a non-linear observation model implements this trait and must be evaluated for a state for which no observation is possible.) Observations with NaN values are treated as missing observations.

fn update( &self, prior: &StateAndCovariance<R, SS>, observation: &Vector<R, OS, Owned<R, OS>>, covariance_method: CovarianceUpdateMethod, ) -> Result<StateAndCovariance<R, SS>, Error>

Given prior state and observation, estimate the posterior state.

This is the update step in the Kalman filter literature.

Implementors