Kalman implementation. More...
#include <Kalman.h>
Public Member Functions  
Kalman (int _n)  
Constructor. More...  
CvMat *  predict (unsigned long tick) 
Predict the Kalman state vector for the given time step This calls updateF for updating the transition matrix based on the real time step. More...  
CvMat *  predict_update (KalmanSensor *sensor, unsigned long tick) 
Predict the Kalman state vector for the given time step and update the state using the Kalman gain. More...  
double  seconds_since_update (unsigned long tick) 
Helper method. More...  
virtual void  update_F (unsigned long tick) 
~Kalman ()  
Destructor. More...  
Public Member Functions inherited from alvar::KalmanCore  
int  get_n () 
Accessor for n. More...  
KalmanCore (const KalmanCore &s)  
Copy constructor. More...  
KalmanCore (int _n)  
Constructor. More...  
virtual CvMat *  predict () 
Predict the Kalman state vector for the given time step . x_pred = F * x. More...  
CvMat *  predict_update (KalmanSensorCore *sensor) 
Predict the Kalman state vector and update the state using the constant Kalman gain. x = x_pred + K* ( z  H*x_pred) More...  
~KalmanCore ()  
Destructor. More...  
Public Attributes  
CvMat *  P 
The error covariance matrix describing the accuracy of the state estimate. More...  
CvMat *  P_pred 
The predicted error covariance matrix. More...  
CvMat *  Q 
The covariance matrix for the process noise. More...  
Public Attributes inherited from alvar::KalmanCore  
CvMat *  F 
The matrix (n*n) containing the transition model for the internal state. More...  
CvMat *  x 
The Kalman state vector (n*1) More...  
CvMat *  x_pred 
Predicted state, TODO: should be protected?! More...  
Protected Member Functions  
void  predict_P () 
Protected Member Functions inherited from alvar::KalmanCore  
virtual void  predict_x (unsigned long tick) 
Protected Attributes  
int  prev_tick 
Protected Attributes inherited from alvar::KalmanCore  
CvMat *  F_trans 
int  n 
Kalman implementation.
The Kalman filter provides an effective way of estimating a system/process recursively (http://www.cs.unc.edu/~welch/media/pdf/kalman_intro.pdf). In this implementation we have separated the Kalman class (KalmanCore, Kalman or KalmanEkf) from the sensor class (KalmanSensorCore, KalmanSensor or KalmanSensorEkf). The selected Kalman class contains always the latest estimation of the process. The estimation can be updated using one or several sensors. This implementation allows SCAAT approach, where there may be several sensors (and several controls) for each Kalman filter (See http://www.cs.unc.edu/~welch/scaat.html).
Currently we have have three levels of implementations for both Kalman and Sensor (core, "normal" and EKF).
The core implementations can be used for really fast barebones core implementations when we have precalculated and constant K. In systems where F, H, Q and R are constants the K will converge into a constant and it can be precalculated. Note, that the core implementation need to assume constant frame rate if F depends on the timestep between frames. Note also, that the coreclasses cannot use EKF Jacobians because they change the H.
The "normal" implementations are used when we have a linear F for Kalman, or linear H for KalmanSensor. The EKF implementations are used when we have nonlinear function f() for KalmanEkf, or nonlinear function h() for KalmanSensorEkf.
Furthermore we have a class KalmanVisualize for visualizing the internal state of Kalman.
Note, that now the KalmanControl is left out from this implementation. But it could be added using similar conventions as the KalmanSensor.
alvar::Kalman::Kalman  (  int  _n  ) 
Constructor.
n  The number of items in the Kalman state vector 
_m  The number of measurements given by this sensor 
Definition at line 161 of file Kalman.cpp.
alvar::Kalman::~Kalman  (  ) 
Destructor.
Definition at line 168 of file Kalman.cpp.
CvMat * alvar::Kalman::predict  (  unsigned long  tick  ) 
Predict the Kalman state vector for the given time step This calls updateF for updating the transition matrix based on the real time step.
x_pred = F*x P_pred = F*P*trans(F) + Q
Definition at line 178 of file Kalman.cpp.

protected 
Definition at line 153 of file Kalman.cpp.
CvMat * alvar::Kalman::predict_update  (  KalmanSensor *  sensor, 
unsigned long  tick  
) 
Predict the Kalman state vector for the given time step and update the state using the Kalman gain.
Definition at line 185 of file Kalman.cpp.
double alvar::Kalman::seconds_since_update  (  unsigned long  tick  ) 
Helper method.
Definition at line 195 of file Kalman.cpp.

virtual 
If your transition matrix F is based on time you need to override this method.
Reimplemented in alvar::KalmanEkf.
Definition at line 174 of file Kalman.cpp.
CvMat* alvar::Kalman::P 
The error covariance matrix describing the accuracy of the state estimate.
CvMat* alvar::Kalman::P_pred 
CvMat* alvar::Kalman::Q 
The covariance matrix for the process noise.