- #1

orochimaru

Hi,

can anyone enlighten me on kalman filter? the maths is overwhelming for me.

Thanks in advance!

can anyone enlighten me on kalman filter? the maths is overwhelming for me.

Thanks in advance!

You are using an out of date browser. It may not display this or other websites correctly.

You should upgrade or use an alternative browser.

You should upgrade or use an alternative browser.

- Thread starter orochimaru
- Start date

- #1

orochimaru

Hi,

can anyone enlighten me on kalman filter? the maths is overwhelming for me.

Thanks in advance!

can anyone enlighten me on kalman filter? the maths is overwhelming for me.

Thanks in advance!

- #2

SGT

Could you post a specific question? What part of the theory you don't understand?

- #3

orochimaru

actually i'm trying to do tracking of a number of corner features across a sequence of image frames.

however, some of these features will be outside the frame after a certain number of frames. Also, some of the corners (which i use opencv to detect) may disappear in between the sequence of frames.

how should i use kalman filter in this case?

i deeply appreciate any help anyone can give because i'm really lost in how to implement this. i mean how should i set the kalman filter equations?

however, some of these features will be outside the frame after a certain number of frames. Also, some of the corners (which i use opencv to detect) may disappear in between the sequence of frames.

how should i use kalman filter in this case?

i deeply appreciate any help anyone can give because i'm really lost in how to implement this. i mean how should i set the kalman filter equations?

Last edited by a moderator:

- #4

SGT

[tex]x_{k+1} = x_k + v_{xk}T

y_{k+1} = y_k + v_{yk}T[/tex]

and the observations under the form:

[tex]z_{k+1} = f(x_{k+1}, y_{k+1}) + v_{k+1}[/tex]

Where [tex]x_k[/tex] and [tex]y_k[/tex] are the coordinates of the point in the instant [tex]t = kT[/tex], [tex]z_{k+1}[/tex] is a function of the coordinates [tex]x_{k+1}, y_{k+1}[/tex] and [tex]v_{k+1}[/tex] is the measurement noise?

If this is possible, you can use a Kalman filter (if the function [tex]f[/tex] is linear), or the Extended Kalman filter (if [tex]f[/tex] is nonlinear)

- #5

orochimaru

1. so is this a nonlinear system?

2. i do not have the Vxk if it refers to the velocity.

3. i have a 1000 of points to track. is this implementable?

btw, is there any such codes already available in the web? i havent managed to find any on the web so far.

Thanks for the help.

- #6

SGT

Not necessarily. The equations of motion are linear, What is your measurement equation? Do you measure the x and y positions directly or do you measure angle and distance to some point?orochimaru said:my mobile robot may be moving in a straight line, could also rotate and also moving in an arc manner (in situation that it travels along a bend).

1. so is this a nonlinear system?

No problem, the filter estimates the components of velocity for you.2. i do not have the Vxk if it refers to the velocity.

Yes.3. i have a 1000 of points to track. is this implementable?

Each implementation of the Kalman filter depends on the dynamic and measurement equations, so I don't believe you could find an implemented filter that suits your application.btw, is there any such codes already available in the web? i havent managed to find any on the web so far.

You're welcome.Thanks for the help.

- #7

orochimaru

- #8

SGT

Share: