Register to reply

Question concerning the extended kalman filter

by Anonymous123
Tags: kalman filter
Share this thread:
Anonymous123
#1
Apr19-13, 07:03 AM
P: 1
Good day,

i read a lot about the kalman filter and the extended kalman filter, but some things are still not clear to me. E.g. I have one question concerning the jacobian matrix of the measurement matrix h. I want to point out my problem with a concrete example:

A vehicle is represented by the following state vector: [tex]x=\begin{pmatrix} x \\ y \\ \varphi \\ v \end{pmatrix}[/tex] (position, rotation and speed).

The equations of the motion model are the following ones:
[tex]
x_x = x_x + x_v \cdot sin(x_\varphi);
[/tex]
[tex]
x_y = x_y + x_v \cdot cos(x_\varphi);
[/tex]

x y phi and v can every second be measured with a failure.

To remind Extended Kalman Filter: http://www.embedded-world.eu/fileadm...Armbruster.pdf (Slide 9)


Question: As visible on the slide 9, I have to calculate the jacobian matrix H for my measurement-function h. The slide points that very well out, in the correction step is H the jacobian matrix and h is my measurement function.

So if I want to consider all elements of the measurement vektor, h would be in my opinion the following matrix:

[tex]h = \begin{pmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} [/tex]

because

[tex]z_k = h*x_t = \begin{pmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} * \begin{pmatrix} x_{t,x} \\ x_{t,y} \\ x_{t,\varphi} \\ x_{t,v} \end{pmatrix} [/tex]

(x_t is the current measurement vector)
Therefore [tex]z_k = x_t[/tex]

But in that case the jacobian matrix [tex]J(h)=H[/tex] becomes

[tex]H = \begin{pmatrix} 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \end{pmatrix} [/tex]

And this means, that the whole correction step of the EKF does not work, because the Kalman gain-matrix also becomes a zero matrix (See again slide 9, there are some matrix multiplications where H is used. When H contains just zeros the gain [tex]K_K[/tex] also becomes zero).

So to conclude: I think I haven't understood the meaning of h and how to calculate H. I appriciate any help and apologize for my english, because Im not a native speaker :)
Phys.Org News Partner Mathematics news on Phys.org
'Moral victories' might spare you from losing again
Fair cake cutting gets its own algorithm
Effort to model Facebook yields key to famous math problem (and a prize)
Stephen Tashi
#2
Apr20-13, 12:52 AM
Sci Advisor
P: 3,252
http://www.google.com/url?sa=t&rct=j...DKnU5Q&cad=rja

Quote Quote by Anonymous123 View Post
So if I want to consider all elements of the measurement vektor, h would be in my opinion the following matrix:

[tex]h = \begin{pmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} [/tex]
In the treatments of the extended Kalman filter that I can read (such as http://www.google.com/url?sa=t&rct=j...DKnU5Q&cad=rja
), [itex] h [/itex] is a vector valued function, not a matrix. In you example it would be the function [itex] h(x,y,\varphi,v) = (x,y,\varphi,v) [/itex]. So the Jacobian of this function is not found by differentiating constants.


Register to reply

Related Discussions
Data fusion using extended kalman filter Set Theory, Logic, Probability, Statistics 8
Extended Kalman Filter : Measurement Matrix Differential Equations 24
A question about Extended Kalman Filter General Math 4
The Extended Kalman Filter Mechanical Engineering 4
Understanding the Extended Kalman Filter Astronomy & Astrophysics 17