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
Heat distributions help researchers to understand curved space
Professor quantifies how 'one thing leads to another'
Team announces construction of a formal computer-verified proof of the Kepler conjecture
Stephen Tashi
#2
Apr20-13, 12:52 AM
Sci Advisor
P: 3,297
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