E[nfc 0 E[VkV

From these conditions it follows that /k has mean

and variance-covariance

P^ = E[(/3fe - /k)(/k - /Jj ] = E[/3fe^T] = E[(yfe - HfcXfc)(yfe - -HfeXfe)T] = E[(efc- ffkr/fcXek- Hiknk)T] P^ = Rk + HHkPk Hf • (4.7.34)

Hence, for a large prediction residual variance-covariance, the Kalman gain

will be small, and the observation will have little influence on the estimate of the state. Also, large values of the prediction residual relative to the prediction residual standard deviation may be an indication of bad tracking data and hence may be used to edit data from the solution.

0 0

Post a comment