- #1
steven_mx
- 2
- 0
Hi,
I'm currently implementing a visual 3D model-based vehicle tracking system as my undergrad dissertation. I've implemented the vehicle localization algorithm and now have an estimate of the x, y location on the ground plane as well as the orientation angle of the vehicle which would like to track using an extended Kalman filter.
I have read the paper by Lou et al. which can be found at http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.132.8981&rep=rep1&type=pdf This should have pretty much what I need. However, I'm having problems understanding how to get the matrices F and H from f(X) and h(X) respectively. I also read the paper at http://academic.csuohio.edu/simond/pubs/ESDNonlinear.pdf which seems to be very similar to what I am implementing, however I'm getting funny results, notably:
- When the vehicle does a turn, the model orientation (predicted by Kalman filter) keeps going round and round in circles instead of following the vehicle motion.
To understand what I mean, take a look at what I've been doing so far. This is before I apply the EKF (my estimates from the vehicle localization module): and this is after I apply the EKF:
I am also unsure about the initial values I should set for the measurement and error covariances, Vk and Qk, so I'm not sure whether the problem is with the way I derived F and H or whether I got the initial values for the covariances wrong.
I have already spent 2 weeks on this and have run out of ideas. I would really appreciate if someone could give me an idea on how to make it work. Thanks a lot in advance.
Steven
I'm currently implementing a visual 3D model-based vehicle tracking system as my undergrad dissertation. I've implemented the vehicle localization algorithm and now have an estimate of the x, y location on the ground plane as well as the orientation angle of the vehicle which would like to track using an extended Kalman filter.
I have read the paper by Lou et al. which can be found at http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.132.8981&rep=rep1&type=pdf This should have pretty much what I need. However, I'm having problems understanding how to get the matrices F and H from f(X) and h(X) respectively. I also read the paper at http://academic.csuohio.edu/simond/pubs/ESDNonlinear.pdf which seems to be very similar to what I am implementing, however I'm getting funny results, notably:
- When the vehicle does a turn, the model orientation (predicted by Kalman filter) keeps going round and round in circles instead of following the vehicle motion.
To understand what I mean, take a look at what I've been doing so far. This is before I apply the EKF (my estimates from the vehicle localization module): and this is after I apply the EKF:
I am also unsure about the initial values I should set for the measurement and error covariances, Vk and Qk, so I'm not sure whether the problem is with the way I derived F and H or whether I got the initial values for the covariances wrong.
I have already spent 2 weeks on this and have run out of ideas. I would really appreciate if someone could give me an idea on how to make it work. Thanks a lot in advance.
Steven
Last edited by a moderator: